Information processing method and system

Abstract

Embodiments of the present specification provide an information processing method and system, wherein the information processing method comprises: constructing a query portrait in response to a query request, and determining a target content portrait associated with the query request in a content portrait library according to the query portrait; generating a query result according to the target content portrait as a response to the query request; determining query record information corresponding to the query request, and storing the query record information to a trusted attribute knowledge base; and updating a candidate content portrait in the content portrait library based on the query record information in the trusted attribute knowledge base in a case where the content portrait library satisfies a verification condition.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to information processing methods and systems. Background Technology

[0002] With the development of computer and internet technologies, users have increasingly higher requirements for the timeliness of content retrieval. Current content retrieval systems generally employ inverted index recall, vector space model recall, multimodal feature recall, and re-ranking mechanisms to determine the relevance of queries to content. With the rapid development of large-scale language models and multimodal large-scale language model technologies, systems can perform structured semantic parsing of user query requests and content documents, generating query profiles and content profiles. The query profile serves as a real-time structured representation of the user query, while the content profile serves as a pre-constructed structured representation of the content document. Profiles play a crucial role in candidate content filtering, attribute weighting, and ranking optimization. However, the reliability of profiles is flawed. During the generation process, content profiles are susceptible to interference from multimodal large-scale language models, image semantic understanding biases, missing content description information, or limitations in attribute extraction algorithms, leading to distorted profile attributes and consequently reducing the overall retrieval accuracy and response efficiency of the system. Therefore, an effective solution is urgently needed to address these issues. Summary of the Invention

[0003] In view of this, embodiments of this specification provide information processing methods. One or more embodiments of this specification also relate to information processing apparatus, information processing system, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0004] According to a first aspect of the embodiments of this specification, an information processing method is provided, comprising: In response to a query request, a query profile is constructed, and based on the query profile, a target content profile associated with the query request is determined in the content profile library. Query results are generated based on the target content profile, serving as a response to the query request; Determine the query record information corresponding to the query request, and store the query record information in a trusted attribute knowledge base; If the content profile database meets the verification conditions, the candidate content profiles in the content profile database are updated based on the query record information in the trusted attribute knowledge base.

[0005] According to a second aspect of the embodiments of this specification, another information processing method is provided, including: In response to an object query request, an object query profile is constructed, and the target object content profile associated with the object query request is determined in the object content profile library based on the object query profile. A list of target objects is generated based on the content profile of the target objects, which serves as a response to the object query request; Determine the query record information corresponding to the object query request, and store the query record information in a trusted attribute knowledge base; If the object content profile library meets the verification conditions, the candidate object content profiles in the object content profile library are updated based on the query record information in the trusted attribute knowledge base.

[0006] According to a third aspect of the embodiments of this specification, an information processing system is provided, including a user terminal and a server, comprising: The server is configured to construct a query profile in response to a query request submitted by the user terminal, and determine the target content profile associated with the query request in the content profile library based on the query profile; generate query results based on the target content profile, and feed the query results back to the user terminal; The user terminal is used to display the query results on the management page; The server is also configured to determine the query record information corresponding to the query request and store the query record information in a trusted attribute knowledge base; and, if the content profile database meets the verification conditions, update the candidate content profiles in the content profile database based on the query record information in the trusted attribute knowledge base.

[0007] According to a fourth aspect of the embodiments of this specification, an information processing apparatus is provided, comprising: The construction module is configured to build a query profile in response to a query request, and determine the target content profile associated with the query request in the content profile library based on the query profile; The generation module is configured to generate query results based on the target content profile, as a response to the query request; The determination module is configured to determine the query record information corresponding to the query request and store the query record information in a trusted attribute knowledge base; The update module is configured to update the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base, provided that the content profile library meets the verification conditions.

[0008] According to a fifth aspect of the embodiments of this specification, another information processing apparatus is provided, comprising: The profile building module is configured to build an object query profile in response to an object query request, and determine the target object content profile associated with the object query request in the object content profile library based on the object query profile. The list generation module is configured to generate a list of target objects based on the content profile of the target objects, as a response to the object query request; The information determination module is configured to determine the query record information corresponding to the object query request and store the query record information in a trusted attribute knowledge base; The profile update module is configured to update the candidate object content profiles in the object content profile library based on the query record information in the trusted attribute knowledge base, provided that the object content profile library meets the verification conditions.

[0009] According to a sixth aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above-described information processing method.

[0010] According to a seventh aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the information processing method described above.

[0011] According to an eighth aspect of the embodiments of this specification, a computer program product is provided, including a computer program or instructions that, when executed by a processor, implement the steps of the information processing method described above.

[0012] The information processing method provided in this embodiment constructs a query profile in response to a query request, and determines the target content profile associated with the query request in a content profile library based on the query profile; generates query results based on the target content profile as a response to the query request; determines the query record information corresponding to the query request, and stores the query record information in a trusted attribute knowledge base; when the content profile library meets the verification conditions, updates the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base, thereby effectively solving the problems of decreased retrieval accuracy and reduced response efficiency caused by the reliability defects of the profiles in the prior art by dynamically correcting the attribute information in the content profile library and optimizing the management of query records, thus effectively improving the accuracy of content retrieval and the system response efficiency. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an information processing method provided in one embodiment of this specification; Figure 2 This is a schematic diagram of information processing in an information processing method provided in one embodiment of this specification; Figure 3This is a flowchart illustrating the processing procedure of an information processing method provided in one embodiment of this specification. Figure 4 This is a flowchart of another information processing method provided in one embodiment of this specification; Figure 5 This is a schematic diagram of the structure of an information processing system provided in one embodiment of this specification; Figure 6 This is a schematic diagram of the structure of an information processing device provided in one embodiment of this specification; Figure 7 This is a schematic diagram of the structure of another information processing device provided in one embodiment of this specification; Figure 8 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0014] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0015] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0016] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, the information should not be limited to the terms. Terms are used only to distinguish information of the same type from one another. For example, first may also be referred to as second, and similarly, second may also be referred to as first, without departing from the scope of one or more embodiments of this specification. Depending on the context, the word “if” as used herein may be interpreted as “when”, “when”, or “in response to a determination.”

[0017] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0018] The technical solutions provided in this application can employ deep learning models with relatively large parameter scales. However, this large model is merely an example; this application does not limit the number of model parameters supported by the deep learning model used, aiming to meet actual needs. The deep learning models involved in this application can be artificial intelligence-based language models (LM) or multimodal models (MM). To address the aforementioned technical problems, this specification provides an information processing method. One or more embodiments of this specification also relate to an information processing apparatus, an information processing system, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0019] In practical applications, the online construction of query profiles is susceptible to interference from users' colloquial expressions, implicit constraints, and long-tail queries, while the offline construction of content profiles is prone to the illusion of multimodal large language models, image understanding bias, incomplete content representation, and insufficient attribute extraction capabilities. Query profiles express the category, intent, constraints, scenario, price, desired effect, and weighted attributes of a query, while content profiles express the category, intent, attributes, scenario, target audience, confidence level, and pending verification status of a document. When profile construction is inaccurate, false matches occur frequently, candidate noise is amplified, and the pressure on the re-ranking model increases, leading to a decrease in the accuracy and efficiency of the retrieval system. Therefore, an effective solution is urgently needed to address these issues.

[0020] In view of this, the information processing method provided in this embodiment constructs a query profile in response to a query request, and determines the target content profile associated with the query request in the content profile library based on the query profile; generates query results based on the target content profile as a response to the query request; determines the query record information corresponding to the query request, and stores the query record information in a trusted attribute knowledge base; when the content profile library meets the verification conditions, updates the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base, thereby effectively solving the problems of decreased retrieval accuracy and reduced response efficiency caused by the reliability defects of the profiles in the prior art by dynamically correcting the attribute information in the content profile library and optimizing the management of query records, thus effectively improving the accuracy of content retrieval and the system response efficiency.

[0021] See Figure 1 , Figure 1 A flowchart of an information processing method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0022] Step S102: In response to the query request, construct a query profile and determine the target content profile associated with the query request in the content profile library based on the query profile.

[0023] Step S104: Generate query results based on the target content profile, as a response to the query request.

[0024] Step S106: Determine the query record information corresponding to the query request and store the query record information in the trusted attribute knowledge base.

[0025] Step S108: If the content profile library meets the verification conditions, update the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base.

[0026] The information processing method provided in this embodiment can be applied to any query scenario, such as object query scenarios (e.g., querying products on a shopping platform, querying videos, texts, and music on an information browsing platform, querying papers on an academic platform), data query scenarios (e.g., querying project documents within an enterprise), or archive query scenarios (e.g., querying target archives maintained on an online platform in an archives). It is used to efficiently and quickly complete query operations, while maintaining a more accurate content profile database to improve query accuracy.

[0027] This embodiment uses the information processing method in the product query scenario as an example for illustration. The descriptions in other scenarios can be found in the same or corresponding descriptions in this embodiment, and will not be elaborated on in this embodiment.

[0028] Specifically, a query request refers to an instruction issued by a user to an information processing system, aimed at obtaining specific information or completing a specific operation. It can take the form of text, voice, or other data input. A query profile refers to a structured data representation obtained after online understanding of a query request. It expresses information such as the query category, intent, constraints, scenario, price, desired effect, and weighted attributes. A content profile library refers to a database storing a large number of candidate content profiles. Each candidate content profile is a structured data representation obtained after offline understanding of a content document, expressing information such as the content category, intent, attributes, scenario, target object, confidence level, and pending verification status. A target content profile refers to one or a group of content profiles in the content profile library that are highly correlated with a specific query request and selected by the system to generate query results.

[0029] Correspondingly, query results refer to the information or operation list generated and returned to the user by the system based on the target content profile, serving as a response to the query request. Query record information refers to the comprehensive data recorded by the system for each query request, including the query profile, target content profile, query process information, confidence level, conflict records, and correction information. The trusted attribute knowledge base refers to the knowledge base used to store and manage query record information. It contains attribute patterns, trusted attribute values, evidence chains, confidence levels, conflict records, and correction rules for reuse in subsequent retrieval and verification tasks. Verification conditions refer to the preset conditions that trigger the update operation of candidate content profiles in the content profile library. When the state of the content profile library or the attributes of its content meet specific criteria, the condition is considered satisfied. Candidate content profiles refer to the structured content representations stored in the content profile library that have not yet been identified as target content profiles but are related to the query request.

[0030] Based on this, after a query request is received, it can be initially analyzed using Natural Language Processing (NLP) technology to extract information such as keywords and entities. Subsequently, the information is mapped to predefined attribute dimensions to construct a query profile. For example, for a text query "finding a red dress priced under 500 yuan," the system can extract dimensions such as "price," "color," and "category" and populate the corresponding values. After constructing the query profile, the system can traverse all candidate content profiles in the content profile library and calculate the matching degree between the query profile and each candidate content profile. The matching degree can be determined through simple keyword matching, attribute value comparison, or rule-based logical judgment. Candidate content profiles with a matching degree reaching a certain threshold are selected, and the one or group with the highest matching degree is determined as the target content profile. For example, if the query profile contains "red dress," the system will search the content profile library for all candidate content profiles containing the attributes "red" and "dress."

[0031] Furthermore, once the target content profile is determined, the system retrieves the corresponding objects from a pre-defined object set based on the structured information contained in that profile. For example, if the target content profile indicates "a red dress from a certain brand," the system will find the actual product matching that description in the product database. Subsequently, the objects are organized into a list or card format and sent to the user's terminal as a response to the query request. This process can generate and present results based solely on the direct attributes of the target content profile.

[0032] Furthermore, after the query request is processed, the system collects the data generated during the query process. This includes the user's original query, the constructed query profile, the final target content profile, intermediate steps involved in the query process (e.g., matching algorithms, filtering rules), the system's confidence assessment of the matching results, records of attribute conflicts, and any manual or automatic correction attempts. This information is integrated into a structured query record. Subsequently, this query record is written into a trusted attribute knowledge base for use in subsequent retrieval or verification tasks. For example, the system can record a user query for "Beijing attractions," with a query profile of "City: Beijing, Type: Attractions," a target content profile of "Forbidden City," and record the confidence level of this match.

[0033] Finally, the system periodically or when triggered by specific events checks whether the content profile library meets preset verification conditions. For example, a simple condition can be set, such as "a content profile exists in the content profile library that has been queried and matched multiple times in the past period, but has never been selected as the final target content profile." Once the verification condition is met, the system retrieves query records related to that condition from the trusted attribute knowledge base. Based on historical query records, the system can identify attribute errors or deviations in candidate content profiles in the content profile library. For example, if the "color" attribute of a content profile is queried as "blue" multiple times, but its actual attribute is "green," the system can mark that attribute as needing correction. Subsequently, the system can adjust or mark the relevant attributes in the candidate content profile according to preset rules to initiate subsequent offline review or correction processes.

[0034] For example, suppose user A sends a query through an e-commerce platform: "I want to find a lightweight T-shirt suitable for summer, in a light color, and priced under 100 yuan." After receiving user A's query, the system analyzes it. For instance, through keyword extraction and semantic analysis, the system can identify information such as "summer," "lightweight," "T-shirt," "light color," and "under 100 yuan." This constructs a structured query profile, including: category "T-shirt," applicable season "summer," style "lightweight," color preference "light," and price range "0-100 yuan." Subsequently, the system uses this query profile to determine the target content profile related to user A's query request in the content profile database. For example, the system will match this query profile with millions of candidate content profiles stored in the content profile database. The candidate content profiles in the content profile library are built offline based on information such as product detail pages, images, and user reviews. For example, a candidate content profile might be described as "Product ID: 12345, Category: T-shirt, Applicable Season: Summer, Style: Pure Cotton, Color: Off-white, Price: 89 yuan". The system calculates the similarity between the query profile and the candidate content profiles, filters out candidate content profiles that highly match user A's intent, and ultimately determines one or a group of content profiles with the highest matching degree as the target content profile. Next, the system generates query results based on the target content profile, serving as a response to user A's query request. For example, if the determined target content profile corresponds to the off-white pure cotton T-shirt with product ID 12345, the system will retrieve the product's detailed information from the product database based on the target content profile, organize it into a product list or product card, and send it to user A's terminal device as the query result.

[0035] Based on this, the system determines the query record information corresponding to User A's query request and stores this query record information in the trusted attribute knowledge base. After User A receives the query results, the system records the complete context information of this query, including User A's original query text, the constructed query profile, the final determined target content profile (e.g., the profile of the off-white T-shirt with product ID 12345), the confidence level of this match (e.g., the system considers the match to be 0.9), and any conflicts or correction information that occur during the query process. The data is packaged into a query record and persistently stored in the trusted attribute knowledge base. This knowledge base not only stores the records but also contains metadata such as attribute definitions and trusted value ranges. Finally, if the content profile library meets the verification conditions, the candidate content profiles in the content profile library are updated based on the query record information in the trusted attribute knowledge base. For example, suppose that over a period of time, the system discovers a candidate content profile (e.g., a T-shirt with product ID 67890) in the content profile library, whose attribute description is "color: dark blue." However, in the trusted attribute knowledge base, numerous query records show that when users search for "light-colored T-shirts," the T-shirt with product ID 67890 is frequently recalled, but ultimately rarely clicked or purchased. This indicates that the "color" attribute of the T-shirt with product ID 67890 is biased, meaning the content profile library is considered to meet the verification conditions. In this case, the system will, based on the query records related to product ID 67890 in the trusted attribute knowledge base, determine that its "color" attribute has an attribute extraction error or attribute offset. Therefore, the system can adjust the confidence level of the "color: dark blue" attribute in the content profile of product ID 67890 and add a verification marker to it. During the offline phase, the evaluation model will identify the verification markers and trigger manual verification or more complex model analysis, ultimately correcting and updating the "color" attribute in the content profile. For example, the color attribute may be adjusted to "dark blue (lighter)" or a "visual light color tendency" attribute may be added, thereby improving the accuracy of future query matching.

[0036] In summary, by dynamically correcting the attribute information in the content profile database and optimizing query record management, the problems of decreased retrieval accuracy and reduced response efficiency caused by the reliability defects of profiles in existing technologies are effectively solved, thereby effectively improving the accuracy of content retrieval and the system response efficiency.

[0037] Furthermore, query requests often possess multi-dimensional and complex characteristics. Simple matching alone is insufficient to accurately capture the user's true intent, leading to a semantic gap between the constructed query profile and candidate content profiles in the content profile library, thus affecting the accuracy of target content profile determination. To address this, in this embodiment, the step of constructing a query profile in response to a query request and determining the target content profile associated with the query request in the content profile library based on the query profile includes: Receive query requests and parse them to obtain multi-dimensional query information; Construct a query profile corresponding to the query request based on the multi-dimensional query information; Align the query profile with the candidate content profiles in the content profile library to obtain the target content profile associated with the query request; The candidate content profiles in the content profile library are constructed based on the multi-dimensional content information corresponding to the candidate content.

[0038] Specifically, a query request is an instruction or expression of intent issued by a user to an information processing system, aiming to obtain specific information or perform specific operations. It is presented in various forms such as text, voice, and images. Parsing a query request refers to the process by which the system deeply understands and structures the received raw request, aiming to extract information with clear semantics and structure from unstructured or semi-structured requests. Obtaining multi-dimensional query information ensures that the parsed results are not merely simple keywords, but contain multiple aspects and levels of attributes or features, such as query category, intent, constraints, context, and sentiment. One implementation method is to use Natural Language Processing (NLP) techniques, such as named entity recognition, intent recognition, and slot filling models, to parse text queries and extract dimensional information such as "product type," "price range," "brand," and "color." Another implementation method is to combine multimodal information processing techniques. For example, for query requests containing voice or images, they can first be converted into text or other structured features through voice or image recognition, and then the aforementioned NLP techniques can be used for multi-dimensional parsing.

[0039] Accordingly, the purpose of building a query profile is to transform the user's complex query intent into a standard format that the system can understand and process, enabling efficient matching with the content profile. A profile contains a series of attribute-value pairs, where each attribute represents a dimension, and its value reflects the user's specific needs or preferences in that dimension. One implementation approach is to organize the parsed multi-dimensional query information, such as "product category: mobile phone," "brand: A," "price: 3000-5000 yuan," and "user intent: purchase," into a JSON object or structured vector as the query profile. Another approach is to utilize knowledge graph technology to map multi-dimensional query information onto predefined knowledge graph nodes and relationships, forming a local knowledge subgraph centered on the query, which serves as the query profile.

[0040] Correspondingly, alignment refers to the process of comparing and matching the constructed query profile with multiple candidate content profiles stored in the content profile library. Its core lies in evaluating the similarity or relevance between the query profile and each candidate content profile to identify the content that best satisfies the user's query intent. The target content profile is the one or a group of content profiles that the system determines is most relevant or best matched to the query request after alignment. One implementation approach is rule-based matching, such as defining a series of matching rules, like "the brand attributes in the query profile must be completely consistent with the brand attributes in the content profile," or "the price range in the query profile must include the price in the content profile." Another implementation approach is to use matching models based on machine learning or deep learning, for example, encoding the query profile and candidate content profiles as vectors, and then calculating the cosine similarity between the vectors or using a neural network model to predict matching scores for alignment.

[0041] Correspondingly, the candidate content profiles in the content profile library are constructed based on the multi-dimensional content information corresponding to the candidate content. The content profile library is a database storing a large number of content profiles, each representing a structured description of a specific content entity (such as a product, article, video, etc.). Candidate content profiles are all possible profiles in the content profile library that are relevant to the user's query. The process of constructing profiles is similar to that of query profiles, involving parsing the original content, extracting its multi-dimensional attribute information, and structuring it. Multi-dimensional content information can include the content's category, theme, keywords, attributes, author, publication time, user reviews, etc. One implementation method is to automatically extract and label original content such as product detail pages, article text, and video metadata. For example, information extraction technology can be used to extract attributes such as "color," "size," and "material" from product descriptions and organize them into structured content profiles. Another implementation method is to combine manual labeling and semi-automated tools to define and assign multi-dimensional attributes to the content. For example, operations personnel can tag specific content, or a crowdsourcing platform can be used for attribute labeling, thereby constructing high-quality candidate content profiles.

[0042] Based on this, a multi-dimensional parsing and alignment mechanism is introduced to improve the matching accuracy between query profiles and content profiles. This method first receives the user's query request and performs in-depth parsing to extract multi-dimensional query information from the original request. Multi-dimensional information is a fine-grained, structured expression of the user's intent; for example, it not only identifies the query keywords but also the underlying intent, constraints, and scenarios. Based on this rich multi-dimensional query information, the system further constructs a query profile that comprehensively reflects the user's query intent. Meanwhile, the candidate content profiles in the content profile library are not simply content tags but are pre-constructed based on the multi-dimensional content information corresponding to the candidate content, also possessing rich structured attributes. Subsequently, the system performs refined alignment processing between the constructed query profile and each candidate content profile in the content profile library. This alignment process overcomes simple keyword matching by comprehensively comparing multiple dimensional attributes contained in the profile, thereby more accurately assessing the semantic relevance between the query intent and the content. Through this multi-dimensional, refined alignment process, the system can accurately filter out the target content profile with the highest relevance to the query request from a massive number of candidate content profiles. This solution is closely integrated with the basic steps of the aforementioned information processing method. In the stage of constructing a query profile in response to a query request and determining the target content profile associated with the query request in the content profile database based on the query profile, the accuracy of the query profile and the precision of content profile matching are greatly enhanced by introducing multi-dimensional analysis and alignment processing. This allows the subsequent generation of query results based on the target content profile to provide a response that better reflects the user's true intent, thereby improving the efficiency of the entire information processing method and user satisfaction.

[0043] For example, suppose a user's query is: "I want to buy a phone with a good camera, with a budget between 3,000 and 5,000 yuan." The request is parsed using a natural language processing model. The parsing process identifies multiple dimensions of information, such as: intent (buy), product category (phone), brand (A), feature preference (good camera), and price range (3,000-5,000 yuan). This multi-dimensional query information is then integrated to construct a structured query profile, such as a JSON object: `{"intent": "buy", "category": "mobile phone", "brand": "A", "feature_preference": "good camera", "price_range": "[3000, 5000]"}`. Simultaneously, a content profile library stores a large number of candidate content profiles for mobile phone products. Each profile is also constructed based on multi-dimensional information from detailed parameters of the phone (such as brand, model, camera pixels, processor, price, etc.). For example, a candidate content profile for a certain mobile phone might be: `{"category": "mobile phone", "brand": "A", "model": "** Pro", "camera_score": "excellent", "price": "4999"}`. Next, the system aligns the user's query profile with all candidate content profiles in the content profile library. This alignment can employ multi-attribute matching algorithms, such as calculating the matching score for each dimension (brand, price, feature preference, etc.) and performing a weighted sum. For the "good camera" feature preference, the system will match it with attributes like "camera_score" or "camera pixels" in the content profile. Finally, based on the comprehensive matching score, the system will select one or more candidate content profiles with the highest scores as the target content profile associated with the query request; for example, determining the profile of the "A ** Pro" mobile phone as the target content profile.

[0044] In summary, by analyzing query requests from multiple dimensions, we can capture the user's true intent more comprehensively and meticulously, constructing a high-precision, structured query profile. Based on this, we perform refined alignment processing between this query profile and candidate content profiles, also built upon multi-dimensional content information, achieving a leap from surface-level keyword matching to deep semantic attribute matching. This improves the matching precision and accuracy between the query profile and the content profile, ensuring that the identified target content profile more accurately reflects the user's actual needs. Therefore, it provides a solid foundation for generating high-quality query results, thereby enhancing the intelligence level of the entire information processing system and the user experience.

[0045] In practical applications, due to the large number of candidate content profiles and the complexity of their feature dimensions, direct alignment often leads to excessive computational resource consumption and makes it difficult to accurately filter out high-quality content that highly matches the user's query intent from a massive amount of candidate content, easily introducing noise interference. Therefore, in this embodiment, the process of aligning the query profile with candidate content profiles in the content profile library to obtain the target content profile associated with the query request includes: Calculate the image similarity between the query image and the candidate content images in the content image library, and filter multiple matching content images in the content image library based on the image similarity. The multiple matching content images are filtered according to the preset image filtering information to obtain a set number of initial content images; The target content profile associated with the query request is determined from the set number of initial content profiles.

[0046] Specifically, the similarity between the query profile and candidate content profiles in the content profile library is calculated to quantify the relevance between the query intent and each candidate content. This similarity can be calculated using various techniques. For example, traditional metrics such as cosine similarity, Euclidean distance, and Jaccard similarity can be used, or deep learning models (such as BERT and Transformer) can be employed to map the query profile and candidate content profiles into a high-dimensional vector space, and then the distance or angle between their vectors can be calculated to represent the similarity. In this way, the system can initially identify content that is semantically or attribute-wise similar to the user's query intent.

[0047] Accordingly, filtering multiple matching content images from the content image database based on image similarity is a step to quickly narrow down the candidate range from a massive amount of content images, based on the aforementioned similarity calculation results. For example, a similarity threshold can be set, and all candidate content images with similarity higher than the threshold can be selected; alternatively, the top N (where N is a preset integer) candidate content images with the highest similarity can be selected as the matching content images, sorted from highest to lowest similarity. This step aims to efficiently eliminate a large amount of irrelevant noise, laying the foundation for subsequent refined processing.

[0048] Accordingly, multiple matching content profiles are filtered according to preset profile filtering information to further improve the quality and business relevance of the screening results. Preset profile filtering information may include, but is not limited to: business rules (e.g., content type, publication time, geographical restrictions, user preference tags, etc.), content quality scores, compliance requirements, blacklists and whitelists, etc. The filtering information can serve as a series of rules or conditions to perform secondary verification on the filtered matching content profiles, eliminating content that, although highly similar, does not actually meet specific business logic or quality standards, thereby obtaining a set number of initial content profiles. For example, if a user searches for "restaurants in Beijing," even if a restaurant in Shanghai has a high similarity, if the filtering information includes a rule that "the region must be Beijing," then that Shanghai restaurant will be filtered out.

[0049] Accordingly, obtaining a set number of initial content profiles refers to the final set of content profiles that are retained after similarity screening and filtering, and are of a controllable quantity and high quality. The "set number" can be fixed, such as always retaining 100 initial content profiles; or it can be dynamically adjusted, for example, determined based on the complexity of the query, system load, or subsequent processing capabilities. This step ensures that the subsequent determination of target content profiles is carried out within a relatively small and high-quality set, thereby improving efficiency and accuracy.

[0050] Correspondingly, identifying the target content profile relevant to the query request from a set number of initial content profiles is a crucial step in ultimately determining the content profile that matches the user's query intent. At this stage, more refined and complex ranking or selection models can be employed. For example, the Rerank model can be used to re-rank the initial content profiles, combined with the fusion analysis of real-time user behavior data, multimodal information (such as images and video content), or decision-making based on expert system rules. The system can then accurately identify the target content profile that is most relevant to the query request and best meets the user's needs from the high-quality initial content profiles.

[0051] Based on this, a phased, refined screening mechanism effectively solves the problem of balancing efficiency and accuracy in the matching process of massive content profiles. First, by calculating the similarity between the query profile and candidate content profiles, a candidate set with high relevance to the query intent can be quickly identified from a global perspective. This similarity metric achieves initial dimensionality reduction and coarse screening, ensuring that subsequent processing focuses on potentially relevant content. Second, pre-defined profile filtering information is introduced to perform secondary filtering on the initially selected matching content profiles. The core of this process is to use predefined rules or constraints to eliminate noisy content that, while highly similar, does not meet specific business logic or quality requirements, thereby obtaining a set number of initial content profiles and ensuring the quality of the candidate set. Finally, the final target content profile is determined from the initial content profiles after multiple rounds of filtering. This progressively layered screening logic not only reduces the computational pressure of directly performing complex alignment processing but also enhances the ability to accurately capture query intent through the introduction of a filtering mechanism, thus optimizing both the relevance and accuracy of the final target content profile.

[0052] For example, suppose a user initiates a query request: "Find a restaurant in location A with an average cost of 100-200 yuan per person, suitable for family gatherings." First, the system constructs a query profile based on this query request, including dimensions such as "Location: Location A," "Average cost per person: 100-200 yuan," and "Scenario: Family gathering." Next, the system retrieves a large number of candidate content profiles (e.g., restaurant profiles from various locations) from a content profile library. In the profile similarity calculation stage, the system can use a pre-trained language model to convert the query profile and all candidate content profiles into vector representations, and then calculate the cosine similarity between them. For example, profiles related to "restaurants in location A" will obtain a high similarity. Based on the similarity, the system will filter, for example, the top 500 most similar matching content profiles. Subsequently, the system filters these 500 matching content profiles according to preset profile filtering information. The filtering information includes: the restaurant's operating status must be "open for business," the restaurant's rating must be higher than 4.0, and it must contain the "family gathering" tag, etc. After filtering, the system obtains, for example, 50 initial content profiles. Finally, in the initial content profile, the system can use a more refined ranking model, combining factors such as user history, real-time popularity, and comment keywords (such as "private room"), to ultimately determine 3-5 target content profiles that match the user's query intent, serving as the basis for responding to the query request.

[0053] In summary, by introducing profile similarity calculation for coarse screening, the amount of data to be processed is reduced, improving processing efficiency. Fine filtering using pre-set profile filtering information effectively eliminates noisy content that does not conform to business logic or quality requirements, ensuring the quality of the candidate set. Finally, the target content profile is determined from the high-quality initial content profiles, ensuring the accuracy of the final matching results. This multi-stage, progressive filtering mechanism enables the system to respond to query requests more efficiently and accurately, improving user experience and system performance.

[0054] Furthermore, simply storing query records often lacks records of the deep logical connections between query profiles and content profiles. This results in the system lacking sufficient multi-dimensional attribute structure information, confidence levels, and conflict records as a basis for judgment when performing profile verification or correction, making it difficult to achieve accurate profile quality assessment and optimization.

[0055] In this embodiment, determining the query record information corresponding to the query request and storing the query record information in a trusted attribute knowledge base includes: Determine the query profile corresponding to the query request and the target content profile associated with the query request; Obtain the image attribute structure information and image credibility attribute values ​​corresponding to the query image and the target content image, respectively; Determine the query process information, confidence level, conflict records, and correction information corresponding to the query request; Based on the profile attribute structure information, the profile credibility attribute value, the query process information, the confidence level, the conflict record, and the correction information, construct the query record information corresponding to the query request; The query record information is stored in a trusted attribute knowledge base, wherein the query record information stored in the trusted attribute knowledge base is used to perform retrieval tasks and / or verification tasks.

[0056] Specifically, a query profile is a structured representation of a user's query intent, constraints, and context, while a target content profile is the most relevant content matched from a content profile library based on the query profile. Identifying both is fundamental to building detailed query records; it links the query input and system output, providing a clear context for subsequent profile quality assessment and optimization. One approach is to obtain the generated query profile through the system's internal query processing module and use matching algorithms (such as vector similarity matching and rule matching) to retrieve and determine the most matching target content profile from the content profile library. Another approach is to capture and transmit the identified query profile and the selected target content profile as intermediate results during the query request processing flow. The goal is to obtain the profile attribute structure information and profile credibility attribute values ​​corresponding to the query profile and the target content profile, respectively. This step transforms the abstract profile content into quantifiable and analyzable structured data. The profile attribute structure information defines the dimensions of the profile, the type of each dimension, and the hierarchical relationship. The profile credibility attribute values ​​quantify the reliability or confidence level of the attributes. Information can be obtained by accessing the profile management module or profile storage database to query predefined profile metadata and real-time calculated or stored attribute confidence levels. Alternatively, it can be obtained by calling the profile parsing service to perform structured parsing and credibility assessment on the raw profile data. The next step is to determine the query process information, confidence level, conflict records, and correction information corresponding to the query request. This step aims to record dynamic information and anomalies during the processing of the query request. Query process information describes the complete path from request to response. Confidence level reflects the reliability of the entire query result. Conflict records refer to inconsistencies or contradictions discovered during profile matching or result generation. Correction information records the automatic or manual intervention measures taken by the system to resolve conflicts or errors. Information can be determined by setting up instrumentation points or logging mechanisms in the query processing chain to capture the status and decisions at each stage in real time. Alternatively, a dedicated monitoring module can be used to identify and record abnormal events during the query process and generate corresponding conflict and correction logs. Based on the profile attribute structure information, the profile credibility attribute values, the query process information, the confidence level, the conflict records, and the correction information, query record information corresponding to the query request is constructed. This step integrates the above multi-dimensional information into a unified, structured query record. The construction process can be achieved by defining a standardized data model or data structure, encapsulating various information items according to preset fields and formats. Alternatively, it can be done through data integration services, extracting relevant data from different data sources, cleaning, transforming, and loading it to form a complete query record.The query record information is stored in a trusted attribute knowledge base. This knowledge base is used to perform retrieval and / or verification tasks. This step involves persistently storing the constructed query record information for later use. Storage can be achieved through a relational database, where structured query records are stored as individual records in a table. Alternatively, it can be stored through a NoSQL database or a knowledge graph database, allowing for more flexible storage of complex query records and their relationships, thus supporting efficient retrieval and complex verification analysis.

[0057] Based on this, a comprehensive query record system was established, encompassing multi-dimensional attribute structures, reliable attribute values, query processes, confidence levels, conflict records, and correction information. This system provides a traceable and quantifiable data asset. First, by defining the query profile and the target content profile, the source of the request-content match can be accurately identified, providing a clear context for subsequent analysis. Building upon this, the profile attribute structure information and reliable attribute values ​​are obtained, aiming to transform unstructured profile content into computable attribute dimensions. Reliable attribute values ​​quantify the reliability of the profile, providing fundamental data support for subsequent profile quality assessment. Simultaneously, defining query process information, confidence levels, conflict records, and correction information allows for a complete record of the query execution path and abnormal states. In particular, the introduction of conflict records and correction information enables the system to capture deviations in the profile during practical applications, providing feedback signals for subsequent automatic correction and model optimization. Finally, based on the aforementioned multi-dimensional information, query records are constructed and stored in a trusted attribute knowledge base. This represents a shift from single query records to a structured knowledge base, enabling retrieval and verification tasks to perform in-depth analysis based on high-value record information. This effectively improves the system's ability to control profile quality and solves the problem of uncontrollable profile quality caused by the lack of feedback mechanisms in practical applications. This solution is closely integrated with the steps described above: responding to query requests to construct query profiles and determining the target content profile related to the query request in the content profile library based on the query profile. After the system has successfully matched the target content profile based on the query profile, it can delve deeper, moving beyond simply focusing on the matching result to deeply analyze and record the quality of the matching process and results. By structurally integrating key information such as query profiles, target content profiles, their attribute structures, credibility, and the entire query process's flow, confidence level, potential conflicts, and corrective actions, and storing this information in the trusted attribute knowledge base, the system can provide extremely rich and detailed evidence for subsequent profile verification and updates. This deep recording mechanism enables the system not only to know "what was matched", but also "why it was matched", "whether the match was good or bad", and "where the problem occurred", thus providing strong data support for subsequent profile optimization and system iteration, improving the reliability of the profile and the intelligence level of the system.

[0058] For example, when searching for restaurants with a rating of 4.5 stars or higher near region B, the system first responds by constructing a corresponding query profile. This profile includes attributes such as "Intent: Restaurant Search," "Location: Region B," and "Rating: 4.5 stars or higher." Next, the system uses this query profile to determine the associated target content profile in the content profile library. For instance, a restaurant's content profile might include attributes such as "Name: abcd," "Address: Block C, Region B," and "Rating: 4.7 stars." After determining the query profile and target content profile, the system further obtains their respective profile attribute structure information and profile confidence attribute values. For example, the attribute structure information for "Location: Region B" in the query profile indicates a geographic location entity, with a confidence attribute value of 0.95, indicating a high level of confidence in the system's identification of this location. Similarly, the attribute structure information for "Rating: 4.7 stars" in the target content profile indicates a numerical evaluation metric, with a confidence attribute value of 0.98, indicating that the rating data originates from an authoritative platform and has been verified. Simultaneously, the system determines the query process information, confidence level, conflict records, and correction information corresponding to this query request. The query process information records that the request passed through the "Natural Language Understanding Module," "Profile Matching Module," and "Result Sorting Module." The overall confidence level of this query is 0.90. If, during the matching process, the system finds a conflict between the "Cuisine" in the user's query profile and the "Cuisine" in a candidate content profile, but due to the extremely high matching degree of other attributes, the system still considers it as a second-best option, this conflict record will be recorded. If the system automatically adjusts the weight of a certain attribute after matching to improve the matching degree, the corresponding correction information will be recorded. Next, the system will construct a structured query record information based on the profile attribute structure information, profile credibility attribute values, query process information, confidence level, conflict record, and correction information obtained above. The query record information can be a JSON object containing fields such as query ID, user ID, query profile details, target content profile details, credibility of each attribute, query path, final confidence level, conflict details (such as conflict attribute, conflict value, conflict type), and correction operation (such as attribute value before correction, attribute value after correction, and correction strategy). Finally, the system will store the constructed query record information in a trusted attribute knowledge base.

[0059] In summary, by comprehensively recording the attribute structure and reliable attribute values ​​of both the query profile and the target content profile, as well as the process, confidence level, conflicts, and correction information during the query, the system can construct highly structured and information-rich query records. These records are stored in a reliable attribute knowledge base, providing a solid data foundation for subsequent profile quality assessment, problem diagnosis, and system optimization. Specifically, when a profile deviates, the system can trace back to the specific query scenario, the cause of the conflict, and the correction process, thereby achieving precise location and efficient correction of profile errors, improving the reliability and accuracy of the profile, and ultimately enhancing the quality of query results and user satisfaction.

[0060] Firstly, the content profile database contains a large number of low-quality or redundant content profiles that have not been selected by users. These profiles not only consume storage resources but also generate noise during the retrieval process, leading to inaccurate search results. Existing solutions lack an automatic identification and correction mechanism for unselected profiles, making it difficult to effectively distinguish which profiles were not selected due to attribute extraction errors, attribute offsets, or mismatches with the query intent. This hinders the accurate maintenance and quality improvement of the content profile database. Therefore, in this embodiment, when the content profile database meets the verification conditions, updating the candidate content profiles in the content profile database based on the query record information in the trusted attribute knowledge base includes: Based on the query record information stored in the trusted attribute knowledge base, if it is determined that there is a first content profile in the content profile library that has participated more than a set threshold number of times and has not been selected, then the content profile library is determined to meet the verification conditions. The content attribute information to be corrected is determined in the first content profile, and the first query record information associated with the content attribute information to be corrected is read from the trusted attribute knowledge base; Based on the first query record information, determine whether there is an attribute extraction error in the attribute information of the content to be corrected, or determine whether there is an attribute offset in the attribute information of the content to be corrected, or determine whether the attribute information of the content to be corrected does not match the query intent based on the first query record information; If so, adjust the attribute confidence of the content attribute information to be corrected in the first content profile, and add verification mark information for the content attribute information to be corrected; In the offline phase, the evaluation model is used to identify the review marker information, and the content attribute information to be corrected in the first content profile is reviewed and updated based on the recognition results.

[0061] Specifically, if a content profile in the content profile library is found to have a participation count exceeding a set threshold but is not selected, the content profile library is deemed to meet the verification criteria. This aims to identify content profiles that are frequently displayed during retrieval but have extremely low user selection rates, indicating quality issues. One implementation involves the system continuously recording the exposure count (participation count) and user selection count for each content profile. When a content profile's exposure count exceeds a preset threshold (e.g., 1000 times), but its selection count is zero or significantly below average, that content profile is identified as the "first content profile," and the content profile library is considered to meet the verification criteria. Another implementation involves a periodically running monitoring module that scans all content profiles in the content profile library. This module extracts historical participation data and user behavior data for each content profile from a trusted attribute knowledge base, calculating its participation count and selection rate. If a content profile's participation count exceeds a dynamically set threshold (e.g., based on the average exposure of its category) and its selection rate is below a statistically significant level (e.g., below 1%), it is marked as the "first content profile," triggering the content profile library's verification process.

[0062] Accordingly, the system identifies the content attribute information to be corrected in the first content profile and retrieves the first query record information related to the content attribute information to be corrected from the trusted attribute knowledge base. This aims to accurately locate the specific problematic attributes in the first content profile and obtain relevant historical query data for diagnosis. One implementation involves the system analyzing the internal structure of the first content profile and identifying all its attributes. For each attribute, the system queries the trusted attribute knowledge base, retrieving all query record information related to that attribute, especially those records that resulted in the first content profile being displayed but not selected by the user. Through statistical analysis, the attributes with the highest frequency or greatest impact in non-selection events are identified as "content attribute information to be corrected." Another implementation involves utilizing a machine learning model trained on historical data. This model can predict the "content attribute information to be corrected" most likely to cause the user to not select the content based on the overall performance of the first content profile and the characteristics of its various attributes. Once the attribute is determined, the system extracts all "first query record information" directly related to the performance of that attribute in non-selection queries from the trusted attribute knowledge base.

[0063] Accordingly, based on the first query record information, it is determined whether there are attribute extraction errors in the content attribute information to be corrected, or whether there is attribute shift in the content attribute information to be corrected, or whether the content attribute information to be corrected does not match the query intent. This aims to perform multi-dimensional diagnosis of problematic attributes and find their root causes. To determine if there are "attribute extraction errors," the value of the attribute in the first content profile can be compared with the record in the original content source (e.g., original document text, image metadata) in the first query record information. If there is an inconsistency, it is determined to be an extraction error. To determine if there are "attribute shifts," it is analyzed whether the user's query intent in the first query record information is consistent with the semantics expressed by the attribute. For example, if a user queries "light luxury style" but the attribute is "minimalism," and there are multiple instances of not selecting it, then there is an attribute shift. To determine if there is a "mismatch with query intent," it is assessed whether the attribute fundamentally conflicts with the user's actual query intent when matching the user's query. For example, if the user's query intent is "purchase," but the attribute describes "product review," then there is an intent mismatch. In addition, semantic analysis models or natural language processing techniques can be used to conduct in-depth analysis of the query text, user behavior context, and attribute information of the content to be corrected in the first query record information, so as to identify semantic inconsistencies, conceptual deviations, or disconnections from the user's actual needs, thereby determining the above three types of errors.

[0064] Correspondingly, if so, the system adjusts the attribute confidence score of the content attribute information to be corrected in the first content profile and adds review marker information to the attribute information to be corrected, aiming to immediately reduce the impact of the problematic attribute and mark it as requiring further manual or offline model review. One implementation is that once a problem is diagnosed, the system immediately reduces the numerical confidence score of the "content attribute information to be corrected" in the "first content profile," for example, from 0.9 to 0.5. Simultaneously, a boolean flag (e.g., `needs_offline_review = true`) is set in the attribute's metadata as a "review marker information." Another implementation is that the system dynamically adjusts the magnitude of the confidence score decrease based on the error type and severity. For example, extraction errors cause a significant drop in confidence, while intent mismatch causes a slight drop. Simultaneously, the system generates a unique "review marker information" for this attribute and adds it to the review task queue for offline evaluation model processing.

[0065] Correspondingly, during the offline phase, an evaluation model is used to identify review marker information. Based on the identification results, the content attribute information to be corrected in the first content profile is reviewed and updated. The aim is to ultimately correct problematic attributes through deeper and more professional analysis, ensuring the accuracy of the content profile. One implementation is that during offline periods with low system load (e.g., at night), a batch processing program scans the content profile library and identifies all "content attribute information to be corrected" with "review marker information." The "evaluation model" can be a more complex machine learning model (e.g., an LLM specifically for attribute validation) or a human review platform. This model or platform performs a detailed review of the problematic attributes based on the original content, the first query record information, and other contextual data. Based on the review results, if an attribute error is confirmed, the "content attribute information to be corrected" is corrected, for example, by updating its value, adjusting its weight, or deleting the attribute. Another implementation is that the evaluation model can be a rule-based expert system that combines domain knowledge and pre-defined validation rules to automatically review attributes with "review marker information." Review and update include re-extracting attributes from the original content, adjusting the semantic representation of attributes, or reassigning attribute categories or weights based on the new understanding.

[0066] Based on this, a verification triggering mechanism based on participation frequency and selection rate is introduced to accurately locate inefficient content in the content profile library. By setting a participation frequency threshold, candidate content with a certain exposure but poor conversion effect can be effectively filtered out, thus focusing on the objects that truly need optimization. After identifying the content to be corrected, real query records stored in the trusted attribute knowledge base are compared, the core of which is to associate offline profile attributes with online real user behavior. By judging attribute extraction errors, attribute offsets, or intent mismatches, the root cause of profile quality problems can be diagnosed from multiple dimensions, rather than blindly adjusting. By adjusting attribute confidence and adding verification marks, this application establishes a phased correction process. Adjusting confidence can immediately reduce the weight of erroneous attributes in subsequent searches and reduce noise interference; while adding verification marks provides a clear optimization target for the evaluation model in the offline stage, enabling the evaluation model to conduct in-depth verification and correction of the marked content. This combination of online diagnosis and offline verification ensures the real-time response capability of the retrieval system and ensures the accuracy and stability of profile updates through the intervention of the offline model.

[0067] For example, suppose a user searches for "science fiction movie" on an online video platform. The system constructs a query profile based on the query request and identifies the relevant target content profile in the content profile library, thus generating query results. Simultaneously, the system stores the user's query behavior, query profile, and target content profile as query record information in a trusted attribute knowledge base. Now, suppose there is a video in the content profile library whose original description is "a philosophical exploration of the future world." However, due to an MLLM attribute extraction error, its content profile is incorrectly labeled with a high-confidence attribute: "Type: Action." The video is therefore frequently recommended to users searching for "action movie" (high engagement), but users rarely click to watch it (not selected). At this point, based on the query record information stored in the trusted attribute knowledge base, the system finds that the "engagement count" of this video content profile far exceeds a set threshold (e.g., 1000 times), but its "selection rate" is less than 1%. Therefore, the system determines that this content profile is the "first content profile" and that the content profile library meets the verification conditions. The system further analyzed the "first content profile" and identified the attribute "Type: Action Movie" as "content attribute information to be corrected". Subsequently, the system retrieved the "first query record information" related to this attribute from the trusted attribute knowledge base. This record contained detailed context of the user searching for "action movie" but not selecting the video. Based on the "first query record information", the system compared the original video description "a philosophical exploration of the future world" with the attribute "Type: Action Movie" and determined that there was an "attribute extraction error". Therefore, the system immediately adjusted the attribute confidence of the "content attribute information to be corrected" (i.e., "Type: Action Movie") in the "first content profile" from 0.9 to 0.2, thereby reducing its weight in subsequent online recommendations. Simultaneously, the system added a "verification marker" to this attribute and added it to the offline verification queue. In the offline phase, a specialized evaluation model (e.g., a deep learning model combining video content analysis and text semantic understanding) identified this "verification marker information". The evaluation model conducts an in-depth analysis of the video's original description, the video content itself, and related user feedback. It ultimately confirms that the attribute extraction was indeed flawed and corrects it to "Type: Science Fiction" or "Type: Philosophical Film." Based on the evaluation model's identification results, the system reviews and updates the "Content Attribute Information to be Corrected" in the "First Content Profile," correcting its attribute values ​​to the correct ones and restoring its confidence level.

[0068] In summary, by adjusting attribute confidence online, the negative impact of problematic attributes on search results can be reduced instantly, decreasing false matches and candidate noise. Simultaneously, by adding verification markers and utilizing an offline evaluation model for deep verification and updates, the long-term accuracy and reliability of content profiles are ensured, thereby improving the precision of content retrieval and user experience. This mechanism, combined with fundamental information processing methods, enables the system to continuously and adaptively optimize content profile quality, solving the problem of unreliable profiles leading to poor search results in traditional solutions, and providing users with more accurate and relevant query results.

[0069] Secondly, during the construction and maintenance of the content profile database, attribute conflicts among candidate content make it difficult for the system to accurately determine the authenticity and accuracy of the candidate content profiles, thus affecting the quality of the query results. To address this, in this embodiment, updating the candidate content profiles in the content profile database based on query record information in the trusted attribute knowledge base, when the content profile database meets the verification conditions, includes: Aggregate the candidate content corresponding to the candidate content profile in the content profile library to obtain candidate content clusters; If there are candidate contents with conflicting attributes in the candidate content cluster, it is determined that the content profile library meets the verification conditions; Determine the conflicting content attribute information in the second content profile corresponding to the candidate content with the attribute conflict, and read the second query record information associated with the conflicting content attribute information in the trusted attribute knowledge base; Based on the second query record information, determine whether there are any abnormal items in the conflict content attribute information; If so, adjust the attribute weight of the conflicting content attribute information in the second content profile, and add correction mark information for the conflicting content attribute information; In the offline phase, the evaluation model is used to identify the correction marker information, and the conflict content attribute information in the second content profile is corrected and updated based on the identification results.

[0070] Specifically, aggregation refers to bringing together multiple independent entities with similar characteristics or attributes to form a logical set or group. Here, its function is to identify different candidate content describing the same or similar topics, products, or services in the content profile library, thus laying the foundation for subsequent attribute conflict detection. One implementation method is that the system can use text similarity algorithms or multimodal similarity algorithms to compare candidate content pairwise, grouping content with similarity exceeding a preset threshold into the same candidate content cluster. Another implementation method is to automatically group candidate content with the same or similar classification tags based on a predefined content classification or tagging system, forming candidate content clusters.

[0071] Correspondingly, attribute conflict refers to inconsistencies or contradictions in the description of the same attribute by different candidate content profiles within the same candidate content cluster. Determining that the content profile library meets the verification conditions indicates that the system has identified potential issues requiring further review and correction, thus triggering subsequent update processes. Attribute conflict detection can employ statistical analysis methods. For example, within a content cluster, if most content profiles display one value for a certain attribute, while a few display a different value, a conflict can be considered to exist. Alternatively, a rule engine can be used for detection, with pre-defined business rules that mark inconsistencies as conflicts.

[0072] Correspondingly, conflicting content attribute information refers to the specific attribute and its value that are involved in the conflict after an attribute conflict is detected. "Second query record information" refers to historical query data related to the conflicting attribute, obtained from a trusted attribute knowledge base. Query records contain information such as the user's true query intent for the attribute, query results, and user behavior feedback, possessing high credibility and serving as a basis for judging the authenticity of the attribute. Determining conflicting content attribute information can be achieved by traversing the attributes of all content profiles in the candidate content cluster and comparing their values ​​on specific attributes. Reading the second query record information can be done by searching the trusted attribute knowledge base based on the identifier of the conflicting content attribute information to obtain all relevant historical query records.

[0073] Accordingly, the step of determining whether there are any anomalies in the conflicting content attribute information based on the second query record information aims to verify the accuracy of the conflicting attributes using real user query data. If a conflicting attribute does not match a large number of reliable query records, then the attribute is an "anomaly," meaning erroneous or inaccurate information. One method is to count the number of queries or user feedback in the second query record information that match the conflicting attribute value; if the consistency ratio is lower than a preset threshold, it is judged as an anomaly. Another method is to use a machine learning model, taking the conflicting attribute information and the second query record information as input, to train the model to predict whether the attribute is an anomaly.

[0074] Correspondingly, adjusting attribute weights refers to decreasing or increasing the importance or credibility of a particular attribute in the content profile. For conflicting attributes identified as anomalies, their weights will be reduced to minimize their impact on query results. "Adding correction flag information" provides guidance for manual or automated review and correction of this attribute in subsequent offline stages. This flag can include information such as the conflict type and suggested correction direction. Adjusting attribute weights can proportionally reduce their weight values ​​based on the severity or confidence level of the anomaly. Adding correction flag information can be achieved by adding a field to the content profile metadata to record statuses such as "to be corrected" or "abnormal attribute," and can include a timestamp and detection source.

[0075] Correspondingly, the offline phase refers to the stage where the system processes data and trains models in a non-real-time, non-user-facing environment, executing when the system load is low. This avoids impacting the performance of the online system and allows for more complex calculations and manual intervention. The "evaluation model" is a model specifically designed to process attributes with correction tagging information, and to review and update them. It can be a rule-based system or a more complex machine learning model. "Correction update" refers to the actual modification of conflicting attributes based on the evaluation model's judgment, restoring their accuracy. This includes modifying attribute values, deleting erroneous attributes, or supplementing missing attributes. The evaluation model can be a natural language processing-based entity recognition model used to re-extract attribute values ​​from the original content and compare them with the tagged conflicting attributes. Correction updates can be triggered by a manual review platform, where human reviewers manually modify attribute values ​​based on correction tagging information and the evaluation model's suggestions.

[0076] Based on this, by introducing a candidate content aggregation and attribute conflict detection mechanism, the system achieves automatic identification and closed-loop correction of potential errors in the content profile library. Specifically, the system first aggregates the candidate content corresponding to the candidate content profiles in the content profile library, grouping similar candidate content together to form candidate content clusters. This step aims to gather different content describing the same topic or entity to facilitate the discovery of inconsistencies between them. Once a candidate content cluster is detected to have conflicting attributes, i.e., different content profiles within the same cluster have contradictions or inconsistencies in certain attributes, the system determines that the content profile library meets the verification conditions, thereby triggering subsequent verification and correction processes. On this basis, the system further determines the specific conflicting content attribute information in the second content profile corresponding to the conflicting candidate content. Subsequently, to verify the authenticity of the conflicting attributes, the system reads the second query record information related to the conflicting content attribute information from the trusted attribute knowledge base. The query record information is accumulated based on users' actual query behavior and feedback, and has high credibility. The system will use the second query record information to determine whether the conflicting content attribute information is an anomaly. For example, if an attribute value does not match the actual situation reflected in a large number of user query records, it will be judged as an anomaly. Once a conflicting content attribute is judged as an anomaly, the system will immediately adjust the attribute weight of that conflicting content attribute in the second content profile, reducing its weight to minimize its impact on subsequent query results. Simultaneously, the system will add correction marker information to the conflicting content attribute, explicitly indicating that the attribute needs further review and correction. Information with correction markers will not immediately affect online services but will be pushed to the offline stage for processing. In the offline stage, the system will use an evaluation model to identify the correction marker information and, based on the identification results of the evaluation model, correct and update the conflicting content attribute information in the second content profile. This phased processing approach ensures both the stability and response speed of online queries and, through the offline evaluation and correction mechanism, ensures the accuracy and reliability of the content profile database, thereby effectively solving the retrieval noise problem caused by content profile attribute conflicts and improving the overall quality of the profile database and the accuracy of query results.

[0077] For example, suppose an e-commerce platform's content profile library contains multiple candidate content profiles for "Smartwatch A of a certain brand." These profiles originate from different product detail pages, user reviews, or information uploaded by merchants. The system first aggregates the candidate content for "Smartwatch A," forming a cluster of candidate content for "Smartwatch A." ​​During the aggregation process, the system discovers that most content profiles in the "Smartwatch A" cluster display a "battery life" attribute of "7 days," but one content profile displays "3 days." At this point, the system detects an attribute conflict and determines that the content profile library meets the verification criteria, requiring correction. The system further locates the conflicting content attribute information, namely "battery life: 3 days." To verify the accuracy of the information, the system reads historical query records related to the "battery life" attribute of "Smartwatch A" from a trusted attribute knowledge base. For example, the trusted attribute knowledge base stores a large number of user queries for "How long does Smartwatch A last?" and positive user feedback records for "7-day battery life." Based on the second query record information, the system determines that the "Battery life: 3 days" attribute value is an anomaly because it does not match the actual situation reflected in the vast majority of user queries and feedback. Therefore, the system immediately adjusts the attribute weight of this conflicting content attribute information in the corresponding second content profile, for example, reducing its weight from the default value to 0.2, thus reducing its influence in subsequent queries. Simultaneously, the system adds a "To be corrected" correction mark to the "Battery life: 3 days" attribute. Offline, a dedicated evaluation model (e.g., a rule-based and knowledge graph-based verification system) identifies attributes marked "To be corrected." The evaluation model further queries the brand's official website or authoritative review data to confirm that the actual battery life of "Smartwatch A" is "7 days." Based on the evaluation model's identification results, the system corrects and updates the conflicting "Battery life: 3 days" attribute information in the second content profile, changing it to "Battery life: 7 days," thereby ensuring the accuracy of the content profile database.

[0078] In summary, utilizing real query records from a trusted attribute knowledge base as the verification basis enables objective and reliable judgment of conflicting attributes, avoiding the subjectivity and limitations of single models or manual judgment, and improving the accuracy of attribute verification. Furthermore, by adjusting the weights of abnormal attributes and adding correction markers, the complex correction process is shifted from online real-time processing to an offline stage, ensuring the stability and response efficiency of online queries. Simultaneously, through offline evaluation and model review and updates, continuous optimization and self-correction of the content profiling database are achieved. This not only improves the overall quality and reliability of the content profiling database but also makes the query results generated based on content profiling more accurate, thereby improving user experience, reducing false matching and recall noise, and ultimately enhancing the efficiency and accuracy of the entire information processing system.

[0079] Thirdly, the construction of query profiles often relies on preset weight parameters. If these weight parameters are fixed or lack a targeted dynamic adjustment mechanism, they will be difficult to adapt to subtle changes in user query intent or differences in feature importance in different scenarios, resulting in inaccurate query profile construction and consequently affecting the accuracy of subsequent content profile matching. Therefore, in this embodiment, the method further includes: Determine the behavioral information corresponding to the query record information in the trusted attribute knowledge base, and aggregate the behavioral information; If, based on the aggregation results, a set number of behavioral information items are all associated with the same profile attribute information, the weight parameters corresponding to the profile attribute information are adjusted. These weight parameters are used in the query profile construction phase.

[0080] Specifically, determining the behavioral information corresponding to query records in the trusted attribute knowledge base aims to extract data related to user behavior from the query records stored in the system. Query records are detailed records of the system's query request processing, containing various behavioral data related to user interactions with the system. Behavioral information directly reflects a user's preferences for specific content or attributes, such as user clicks, browsing, purchases, and favorites. This process can be achieved by parsing the query records stored in the trusted attribute knowledge base, identifying and extracting user interaction events, such as which content the user clicked, how long they stayed, and whether they performed subsequent actions, and marking these events as behavioral information. Alternatively, pre-defined behavioral information extraction rules can be used to perform pattern matching on the query records. For example, if the query records contain logs of "clicking content A," then "clicking content A" will be extracted as a behavioral information entry.

[0081] Correspondingly, aggregating the aforementioned behavioral information refers to summarizing, statistically analyzing, and classifying scattered and independent behavioral data to discover underlying patterns, trends, or common preferences. This can eliminate the randomness or noise of individual behaviors, thereby obtaining more representative and statistically significant user behavioral characteristics. This aggregation process can employ a time-window-based aggregation method, for example, counting the number of clicks made by all users on a specific profile attribute within the past 24 hours. Alternatively, it can employ a user group-based aggregation method, for example, summarizing the behaviors of all users interested in a certain type of content to identify the common preferences of that user group.

[0082] Accordingly, when the aggregation results determine that a set number of behavioral information entries are all associated with the same profile attribute information, the judgment is based on the aggregated behavioral information to identify user behavior patterns with high consistency or consensus. Here, "set number" is a preset threshold used to measure the consistency or importance of behavioral information. When the number is reached or exceeded, it indicates that the profile attribute information exhibits significant commonality in user behavior. For example, a small threshold for the number of behaviors can be set; if a profile attribute information is associated with at least N different behavioral information entries (e.g., N clicks), the condition is considered met. Alternatively, a threshold for the proportion of behavioral information association can be set; if a profile attribute information occupies more than P% of the aggregated behavioral information, the condition is considered met.

[0083] Accordingly, adjusting the weight parameters corresponding to the profile attribute information refers to dynamically correcting the importance of the profile attribute information used to construct the query profile based on the identified user behavior patterns. The weight parameters directly affect the contribution of profile attribute information in the query profile construction process. By adjusting the weights, the query profile can more accurately reflect the user's true intent. This adjustment can employ a linear adjustment strategy, for example, increasing the weight parameter of the profile attribute information by a fixed value or a scaling factor when certain conditions are met. Alternatively, a non-linear adjustment strategy can be used, for example, smoothing or accelerating the adjustment of the weight parameters using a sigmoid function or exponential function based on the quantity or intensity of behavioral information.

[0084] Accordingly, the weight parameters are used in the query profile construction stage. This feature clarifies the application scenario of the adjusted weight parameters. That is, when constructing the query profile, the system will use the dynamically adjusted weight parameters to calculate the importance of different profile attribute information, thereby generating a more accurate query profile. When constructing the query profile, the value of each profile attribute information can be multiplied by its corresponding weight parameter, and then a weighted sum can be performed to obtain the final query profile representation.

[0085] Based on this, a closed-loop optimization of the query profile construction process is achieved by introducing a dynamic weight adjustment mechanism based on historical behavior feedback. First, the system determines the behavioral information corresponding to the query records stored in the trusted attribute knowledge base. These query records detail the user's interaction history with the system, including the user's response to the query request, the determination of the target content profile, and subsequent user behavior. Analysis of these records allows for the extraction of the user's actual feedback on specific profile attributes during operation, such as clicking, browsing, or purchasing content related to a specific attribute. Behavioral information is direct evidence reflecting the degree of matching between the user's query intent and the content profile attributes. Second, the system aggregates the extracted behavioral information. The purpose of aggregation is to eliminate the randomness or noise of individual behaviors and identify common preferences or behavioral patterns of users on specific profile attributes from a macro-statistical perspective. For example, by statistically analyzing the common selection of a certain type of attribute content by a large number of users under similar queries, the importance of that attribute to the user's intent can be discovered. Next, the system will judge based on the aggregation results. When it is determined that a set number of behavioral information are associated with the same profile attribute information, that is, when a certain profile attribute information shows high consistency or consensus in a large number of user behaviors, the system will trigger the adjustment of the weight parameters corresponding to that profile attribute information. The core of this process is to use high-frequency and consistent behavioral feedback to correct the query profile construction logic. For example, if a large number of users click on the content of "Brand: A" when querying "mobile phone", then the "Brand: A" profile attribute should be given a higher weight when constructing the query profile. Finally, the adjusted weight parameters will be used in the subsequent query profile construction stage. In this way, the system can more accurately capture user intent based on dynamically changing weight parameters, thereby improving the matching accuracy between the query profile and the target content profile. The dynamic adjustment mechanism solves the problem that traditional static weights cannot adapt to dynamic query needs and enhances the system's adaptive ability to understand user intent. This solution, combined with basic information processing methods, no longer relies solely on preset static weights when constructing query profiles in response to query requests, but can dynamically adjust the weights of profile attributes based on historical user behavior feedback. It can be determined that after identifying the query record information corresponding to the query request and storing it in the trusted attribute knowledge base, the record is not only used for the verification and updating of the content profile library, but further, the behavioral information it contains is used to optimize the construction of the query profile itself. When the system finds that a large number of similar queries all point to a specific profile attribute, it will increase the weight of that attribute in the query profile, so that the subsequent query profile construction can more accurately reflect the user's true preferences. Thus, when determining the target content profile of related query requests in the content profile library, it can more effectively filter out content that highly matches the user's intent, and ultimately generate query results that better meet the user's expectations. The feedback loop mechanism improves the intelligence and adaptability of the entire information processing method.

[0086] For example, suppose an e-commerce platform's information processing system stores a large amount of user query records in its trusted attribute knowledge base. These records include keywords used by users, products clicked, browsing duration, and purchase behavior. First, the system determines the corresponding behavioral information from the trusted attribute knowledge base. For instance, if a user queries "thin and light laptop," and ultimately clicks on and purchases a laptop with "Brand: S," "Screen Size: 13 inches," and "Weight: 1.2kg," then "Brand: S," "Screen Size: 13 inches," and "Weight: 1.2kg" are extracted as behavioral information. Next, the system aggregates this behavioral information. For example, over a period of time, the system aggregates 1000 query records related to "thin and light laptop" and finds that 800 of these records are associated with the "Weight: Under 1.5kg" profile attribute, and users have a higher click-through rate and purchase rate for this type of product than for products in other weight ranges. Then, based on the aggregation results, the system determines cases where a set number of behavioral information entries are associated with the same profile attribute information. In this example, all 800 behavioral data points (far exceeding the preset limit of 500) are associated with the "weight: under 1.5kg" profile attribute. This indicates a high degree of consensus and preference among users when searching for "thin and light laptops" regarding the "weight: under 1.5kg" attribute. In this case, the system adjusts the weight parameter of the "weight" profile attribute when constructing the query profile. For example, the weight parameter of the "weight" attribute can be increased from the initial 0.6 to 0.8. Finally, when the user searches for "thin and light laptops" again, the system will use the adjusted, higher weight parameter of the "weight" attribute during the query profile construction stage. It can be determined that the "weight" attribute will play a more important role in generating query profiles, making the generated query profiles more likely to match laptop content profiles with a weight of under 1.5kg. For example, when calculating the similarity between the query profile and the content profile, if the "weight" attribute of the content profile matches the "weight" attribute of the query profile highly, it will receive a higher overall similarity score due to its higher weight, making it easier to be identified as the target content profile.

[0087] In summary, by continuously extracting and aggregating user behavior information from a trusted attribute knowledge base, the system can perceive changes in user intent and the importance of different profile attributes in real time. When a large number of user behaviors are found to consistently point to a specific profile attribute, the system can promptly adjust the weight of that attribute in the query profile construction, enabling the query profile to more accurately and adaptively capture the user's true needs. This not only improves the accuracy of the query profile, thereby optimizing the matching effect between query requests and target content profiles and reducing false matches and candidate noise, but also enhances the intelligence and adaptability of the entire information processing system in understanding user intent, thus providing users with more personalized and efficient query results.

[0088] Furthermore, after obtaining the target content profile, how to transform the abstract profile information into concrete query results that are perceptible to the user, and ensure that the results accurately respond to the user's query request, remains a challenge in achieving a closed-loop information processing mechanism. Without an effective mechanism for mapping from profile to specific objects, the profile matching results cannot be directly transformed into the business services required by the user, thus affecting the usability of the query and the user experience. Therefore, in this embodiment, generating query results based on the target content profile as a response to the query request includes: Based on the target content profile, determine the target object associated with the query request from the candidate object set; A list of objects is generated based on the target object, and the list of objects is sent to the user terminal corresponding to the query request as a response to the query request.

[0089] Specifically, the target object for the related query request is determined from the candidate object set based on the target content profile. This refers to transforming the abstract, matched, and verified target content profile into a concrete, operable business entity. Specifically, the system can filter and match from a pre-built candidate object set (e.g., a product database, service list, knowledge base entries, etc.) based on the attribute information contained in the target content profile (such as product category, functional characteristics, price range, user intent, etc.). One implementation method is to compare each attribute of the target content profile with the metadata of the candidate objects through attribute tag matching, filtering out objects with highly consistent attributes. Another implementation method is to use semantic embedding technology to map the target content profile and candidate objects to a high-dimensional vector space, and determine the most relevant target object by calculating the similarity between the vectors. Furthermore, a rule-based inference engine can be used to trigger preset business rules based on specific combinations of the target content profile, thereby identifying objects that meet the conditions.

[0090] Accordingly, an object list is generated based on the target objects. The purpose of this step is to organize one or more identified target objects into a structured form that is easy to display and transmit. One implementation is to directly encapsulate the identified target objects (e.g., product ID, service ID) and their core identification information into a list structure. Another implementation is that the system further queries detailed information about the target objects, such as product images, prices, detailed descriptions, user reviews, etc., and formats this rich information into each entry in the list. Target objects can also be sorted, aggregated, or further filtered based on user preferences, historical behavior, or business rules to generate personalized or optimized object lists.

[0091] Correspondingly, the object list is sent to the user terminal corresponding to the query request as a response to the query request. This ensures a closed loop in information processing and effectively delivers the processing result to the user. One implementation method is to send the object list data to the user terminal (such as a mobile application or web browser) in JSON, XML, or other data exchange formats via standard network communication protocols (such as HTTP / HTTPS). Another implementation method is to use push notification services or real-time communication technologies to asynchronously push the object list to the user terminal, ensuring that the user can receive the response in a timely manner. In addition, the user terminal can be allowed to actively initiate a request to obtain the object list through an API interface. This object list, as the final response to the query request, will be directly presented on the user interface for the user to browse, interact with, or perform subsequent operations.

[0092] Based on this, after receiving a query request and undergoing preliminary processing (e.g., constructing a query profile and determining the target content profile), the system obtains a highly refined and validated target content profile. This target content profile accurately expresses the user's query intent, constraints, and preferences. Building upon this, the system first utilizes the rich information contained within the target content profile to precisely filter through a pre-defined set of candidate objects to identify target objects highly relevant to the query request. This effectively solves the disconnect between the profile matching results and the final business presentation, ensuring seamless integration from user intent to specific business entities. Subsequently, based on the identified target objects, the system further generates a structured object list. This list not only includes the basic identifiers of the objects but also integrates their detailed attributes and display information, making it more readable and usable. Finally, this object list is sent to the user's terminal as a direct response to the query request. Through the above processing, the complex profile matching process is transformed into intuitively visible query results for the user, ensuring a high degree of consistency between the response content and the user's query intent. It is worth noting that the above processing is closely integrated with the preceding steps (e.g., the construction of query profiles and the verification and updating of the content profile library as described in claims 1 to 7), forming a collaborative information processing system. The preceding steps, through the refined construction of query profiles and content profiles and the continuous verification and updating of the content profile library, greatly improve the accuracy and reliability of the target content profile. It is precisely based on this highly reliable target content profile that the target object can be more accurately identified from the candidate object set, thereby improving the matching degree of the final query results and user satisfaction. This processing method organically combines the retrieval capability of the profile library with the actual supply capability of the object, enabling the system to construct personalized business feedback in real time based on dynamically generated profiles, thereby improving the response quality and effectiveness of user interaction in actual business scenarios.

[0093] For example, suppose a user enters the query via voice assistant or search box: "My home doesn't get enough light, and the leaves are always turning yellow. Can you recommend some inexpensive, glossy green plants that are easy to care for?" The system first constructs a query profile based on the previous steps. For instance, the category is "plants," the intent is "recommend low-priced, shade-tolerant, and easy-to-care-for plants," and constraints include "poor lighting," "around ten yuan," "glossy green," and "easy to care for." Next, the system identifies the target content profile associated with this query from its content profile library. For example, a content profile described as "pothos, peperomia, shade-tolerant, cheap, easy to care for, and with glossy green leaves." Based on this, the system determines the target objects associated with the query from the candidate object set. Specifically, the system accesses a product database (as the candidate object set), which stores detailed information and attribute tags for various plant products. The system filters specific products that meet the criteria based on attributes such as "plant," "shade-tolerant," "cheap," "easy to care for," and "glossy green" in the target content profile, such as "potted pothos," "lucky bamboo," and "spider plant." The specific product instance is the target object. The system then generates an object list based on the target object. For example, for the target object "potted pothos," the system extracts information such as its product name, main image URL, current price (e.g., 9.9 yuan), and brief description (e.g., "indoor shade-tolerant plant, purifies air, easy to care for") from the product database and formats it into a list item. Similarly, "lucky bamboo" and "spider plant" undergo similar processing. Finally, the system generates an object list containing plant product information. This object list is then sent to the user's terminal. For example, JSON data containing product name, image, price, etc., is sent to the user's mobile app via an API interface. After receiving the data, the user's mobile app displays a list of recommended plants on the screen, with each list item corresponding to a specific product. The user can click to view details, add to cart, etc. The user's abstract query intent is transformed into specific, actionable product recommendations, achieving an effective response to the query request.

[0094] In summary, by establishing a mapping mechanism from target content profiles to specific objects, the system can accurately identify and provide actual business entities that highly match the user's query intent, thereby improving the accuracy and usability of query results. Especially with the refined construction and verification of query and content profiles in the preceding steps, the system ensures that the identified target objects have higher relevance and credibility, avoiding incorrect matching and invalid recommendations due to inaccurate profiles. This not only greatly optimizes the user experience, enabling users to directly obtain specific services or goods closely related to their needs, but also improves the overall efficiency and value of the information processing system, achieving a complete closed loop from understanding user intent to business delivery.

[0095] Furthermore, when constructing a trusted attribute knowledge base to store query record information for performing retrieval and verification tasks, as query requests continuously accumulate, the trusted attribute knowledge base will continuously generate massive amounts of query record information. If effective lifecycle management of this information is not implemented, it will lead to increased storage pressure on the knowledge base, and redundant or expired query record information will interfere with the accuracy of subsequent retrieval and verification tasks. Therefore, in this embodiment, after the step of determining the query record information corresponding to the query request and storing the query record information in the trusted attribute knowledge base is executed, the method further includes: Time detection is performed on the query record information stored in the trusted attribute knowledge base; Based on the time detection results, multiple query records that exceed the set time threshold are identified as needing optimization. Obtain the retrieval information corresponding to the plurality of query record information to be optimized, and determine the query record information to be deleted from the plurality of query record information to be optimized based on the retrieval information; Delete the query record information to be deleted from the trusted attribute knowledge base.

[0096] Specifically, time detection is performed on query records stored in the trusted attribute knowledge base to identify query records that have been stored for a long time and whose value has diminished due to reduced timeliness. One implementation is to attach a storage timestamp to each query record. The system periodically scans the trusted attribute knowledge base, comparing the current time with the timestamp of each record to calculate its storage duration. Another implementation is to manage query records according to the batches in which they are generated or stored, and to set an expected lifespan for each batch. When the batch's lifespan ends, the records it contains are included in the time detection scope.

[0097] Accordingly, based on the time detection results, multiple query records exceeding a set time threshold are identified as needing optimization. This step is used to precisely filter out records that have indeed exceeded the preset timeliness standard from the time-detected records. Specifically, the system compares the storage duration of each query record with a pre-set "set time threshold." For example, if the set time threshold is 90 days, all query records with a storage duration exceeding 90 days will be marked as needing optimization. Furthermore, a dynamically adjustable strategy module can be configured to flexibly set different time thresholds based on business needs or the importance of data types, thereby more precisely filtering out records to be optimized.

[0098] Accordingly, obtaining retrieval information for multiple query records to be optimized aims to supplement the activity or importance data of these records in actual use, enabling a more comprehensive evaluation. One implementation involves the system recording corresponding operation logs in the background each time a query record is retrieved, referenced, or involved in a verification task. These logs include information such as retrieval time, retrieval frequency, and associated tasks. When retrieval information is needed, the system analyzes the log data. Another implementation involves associating a metadata field with the query record information during storage, such as a counter to record the number of times it has been retrieved, or a timestamp to record the time of its last retrieval, thereby directly obtaining its retrieval information.

[0099] Accordingly, based on the search information, the system identifies the query records to be deleted from multiple query records to be optimized. This step is based on a comprehensive judgment of time and usage dimensions to identify truly redundant records that can be deleted. Specifically, the system can set a multi-dimensional filtering rule. For example, in addition to meeting the time threshold, it may also require that the number of searches within a specific time period (such as the last 30 days) is lower than a preset "search frequency threshold," or that its "last search time" is earlier than a preset "search timeliness threshold." Only records that meet both conditions will be identified as query records to be deleted. Alternatively, a lightweight decision model can be used, taking time information, search frequency, and importance of related business as input features, and outputting the deletion priority of each record through model evaluation, thereby determining the records to be deleted.

[0100] Correspondingly, deleting query records to be deleted from the trusted attribute knowledge base is the final operation of data cleanup, aiming to physically or logically remove query records that are no longer needed. One implementation is to perform a standard DELETE operation on query records stored in a database (such as a relational or NoSQL database), removing the identified records from the storage medium. Another implementation is to combine logical and physical deletion. First, the records to be deleted are marked as "deleted," but retained in the knowledge base for a period to handle accidental deletion or auditing needs. After the preset retention period is reached, the physical deletion operation is then performed.

[0101] Based on this, an automated lifecycle management mechanism is introduced to dynamically maintain the query record information stored in the trusted attribute knowledge base. After the query record information corresponding to a query request is stored in the trusted attribute knowledge base, the system periodically performs time checks on the records to identify those that have been stored for a long time. By comparing the storage duration of the records with a preset time threshold, the system can initially filter out query record information that has lost its timeliness and needs to be optimized. On this basis, the system further obtains the retrieval information of the records to be optimized in the actual business scenario, such as the frequency of access or the time of the most recent access. Combining the time dimension and retrieval information, the system can more accurately determine which records are truly redundant or expired data, thereby determining the query record information to be deleted. Finally, the determined records to be deleted will be removed from the trusted attribute knowledge base. This solution is closely integrated with the previously mentioned steps of building query profiles, determining target content profiles, generating query results, and updating the content profile library based on the trusted attribute knowledge base. The trusted attribute knowledge base is an effective support for verifying and optimizing content profiles in the entire information processing flow. If the knowledge base is filled with a large amount of redundant or outdated query records, it will not only increase the storage and processing burden, but more importantly, low-quality data will interfere with subsequent verification tasks, leading to incorrect corrections or inefficient updates to the content profile, thus affecting the accuracy of query results. By continuously cleaning the knowledge base, we ensure that the query records stored within it always maintain high timeliness and relevance, thereby providing a reliable data foundation for the accurate updating of the content profile database, and ultimately improving the accuracy and efficiency of the query results generated by the entire information processing system.

[0102] For example, a trusted attribute knowledge base stores query records of user interactions with a customer service robot. To effectively manage the data, the system can be configured with a background service that runs once daily at midnight. First, this service iterates through all query records in the trusted attribute knowledge base and checks the storage timestamp of each record. The system presets a 90-day time threshold; all query records stored for more than 90 days are marked as pending optimization. Next, for these pending optimization records, the system further queries their corresponding retrieval information. For instance, each query record, in addition to its storage timestamp, maintains a "last access time" field and a "total access count" field. The system checks whether the "last access time" of the pending record is earlier than 30 days ago and whether the "total access count" is less than 5 times. Only query records that simultaneously meet all three conditions—"storage time exceeding 90 days," "last access time earlier than 30 days ago," and "total access count less than 5 times"—are ultimately determined as pending deletion. Finally, the system will perform a database deletion operation, permanently removing the identified query records to be deleted from the trusted attribute knowledge base's database tables. In this way, the trusted attribute knowledge base can periodically clean up old and rarely used records, thus maintaining the knowledge base's simplicity and efficiency.

[0103] In summary, by introducing time detection and retrieval information evaluation mechanisms, intelligent lifecycle management of query records in the knowledge base is achieved, ensuring the timeliness and relevance of data. This not only reduces storage costs and system maintenance burden, but more importantly, by providing a high-quality, non-redundant set of query records, it significantly improves the accuracy and efficiency of subsequent retrieval tasks and content profiling verification tasks performed based on this knowledge base. For example, when correcting or refining content profiling, it avoids introducing new biases by referencing outdated or irrelevant query records, thus enabling the entire information processing system to continuously provide more accurate and reliable query results.

[0104] like Figure 2The schematic diagram illustrates that the information processing method provided in this embodiment can be understood as a retrieval self-correction system. This system enables self-iterative optimization of query profiles, content profiles, and a trusted attribute knowledge base. In offline mode, a content profile corresponding to the content can be pre-built. In the online stage, after a user submits a query, a corresponding query profile can be built. Subsequently, through retrieval and feedback processing, profile alignment, candidate compression, and ranking are achieved. Once a matching content profile is found, the system can provide feedback on objects of interest to the user. For example, if a user's query is "poor lighting at home, low-cost low-light plants," the system can recommend plants like pothos. To ensure query accuracy, an Agentic verification layer and a trusted attribute knowledge base (profile wiki) can perform self-correction. The trusted attribute knowledge base stores all information involved in the query process, including attribute schemas, trusted attribute values, evidence and confidence levels, conflict records, and correction rules. During the verification stage, relevant information can be retrieved from this trusted attribute knowledge base for analysis and use. For example, click feedback verification, similar document consensus verification, and similar query consensus verification can be performed to achieve a complete self-correction closed loop. It effectively solves the problems of decreased retrieval accuracy and reduced response efficiency caused by the reliability defects of portraits in existing technologies, thereby effectively improving the accuracy of content retrieval and the system response efficiency.

[0105] The following is in conjunction with the appendix Figure 3 Taking the application of the information processing method provided in this specification in a product recommendation scenario as an example, the information processing method will be further explained. Among other things, Figure 3 A flowchart illustrating the processing steps of an information processing method according to an embodiment of this specification is shown, specifically including the following steps.

[0106] Step S302: Receive the query request and parse the query request to obtain multi-dimensional query information.

[0107] Step S304: Construct a query profile corresponding to the query request based on multi-dimensional query information.

[0108] Step S306: Calculate the image similarity between the query image and the candidate content images in the content image library, and filter multiple matching content images in the content image library based on the image similarity.

[0109] Step S308: Filter multiple matching content images according to preset image filtering information to obtain a set number of initial content images.

[0110] Step S310: Determine the target content image for the associated query request from the set number of initial content images.

[0111] Step S312: Determine the target object of the related query request from the candidate object set based on the target content profile.

[0112] Step S314: Generate an object list based on the target object, and send the object list to the user terminal corresponding to the query request as a response to the query request.

[0113] Step S316: Determine the query profile corresponding to the query request and the target content profile associated with the query request.

[0114] Step S318: Obtain the image attribute structure information and image credibility attribute value corresponding to the query image and the target content image, respectively.

[0115] Step S320: Determine the query process information, confidence level, conflict records, and correction information corresponding to the query request.

[0116] Step S322: Construct query record information corresponding to the query request based on the profile attribute structure information, profile trust attribute values, query process information, confidence level, conflict records, and correction information.

[0117] Step S324: Store the query record information in the trusted attribute knowledge base, wherein the query record information stored in the trusted attribute knowledge base is used to perform retrieval tasks and / or verification tasks.

[0118] Step S326: If the content profile library meets the verification conditions, update the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base.

[0119] The information processing method provided in this embodiment constructs a query profile in response to a query request, and determines the target content profile associated with the query request in a content profile library based on the query profile; generates query results based on the target content profile as a response to the query request; determines the query record information corresponding to the query request, and stores the query record information in a trusted attribute knowledge base; when the content profile library meets the verification conditions, updates the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base, thereby effectively solving the problems of decreased retrieval accuracy and reduced response efficiency caused by the reliability defects of the profiles in the prior art by dynamically correcting the attribute information in the content profile library and optimizing the management of query records, thus effectively improving the accuracy of content retrieval and the system response efficiency.

[0120] See Figure 4 , Figure 4 A flowchart of another information processing method provided according to an embodiment of this specification is shown, which specifically includes the following steps.

[0121] Step S402: In response to the object query request, construct an object query profile, and determine the target object content profile associated with the object query request in the object content profile library based on the object query profile.

[0122] Step S404: Generate a target object list based on the target object content profile, as a response to the object query request.

[0123] Step S406: Determine the query record information corresponding to the object query request, and store the query record information in the trusted attribute knowledge base.

[0124] Step S408: If the object content profile library meets the verification conditions, update the candidate object content profiles in the object content profile library based on the query record information in the trusted attribute knowledge base.

[0125] The other information processing method provided in this embodiment can be applied to query scenarios on any platform, such as querying products on a shopping platform, querying videos, text, and music on an information browsing platform, and querying papers on an academic platform. It is used to provide users with a list of target objects matching their query needs based on the query results, such as a list of products, videos, or papers, for users to browse and read. Subsequently, the query record information corresponding to the determined request can be stored in a trusted attribute knowledge base. If the object content profile library meets the verification conditions, the candidate object content profiles in the object content profile library are updated based on the query record information in the trusted attribute knowledge base. This achieves dynamic correction of attribute information in the profile library and optimizes query record management, effectively solving the problem of decreased retrieval accuracy and reduced response efficiency caused by profile reliability defects in existing technologies, thereby effectively improving content retrieval accuracy and system response efficiency.

[0126] It should be noted that the second information processing method provided in this embodiment belongs to the same inventive concept as the information processing method described above. Any content not described in detail can be referred to the same or corresponding descriptions in the above embodiments. This embodiment will not elaborate further here.

[0127] Corresponding to the above method embodiments, this specification also provides information processing system embodiments. Figure 5 A schematic diagram of the structure of an information processing system according to one embodiment of this specification is shown. Figure 5 As shown, the information processing system 500 includes a server 510 and a user terminal 520, including: The server 510 is configured to construct a query profile in response to a query request submitted by the user terminal, and determine the target content profile associated with the query request in the content profile library based on the query profile; generate query results based on the target content profile, and feed the query results back to the user terminal; The user terminal 520 is used to display the query results on the business management page; The server 510 is further configured to determine the query record information corresponding to the query request and store the query record information in a trusted attribute knowledge base; and, if the content profile database meets the verification conditions, update the candidate content profile in the content profile database based on the query record information in the trusted attribute knowledge base.

[0128] The above is an illustrative scheme of an information processing system according to this embodiment. It should be noted that the technical solution of this information processing system and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the information processing system, please refer to the description of the technical solution of the information processing method described above.

[0129] Corresponding to the above method embodiments, this specification also provides embodiments of an information processing apparatus. Figure 6 A schematic diagram of the structure of an information processing apparatus according to one embodiment of this specification is shown. Figure 6 As shown, the device includes: The construction module 602 is configured to construct a query profile in response to a query request, and determine the target content profile associated with the query request in the content profile library based on the query profile; The generation module 604 is configured to generate query results based on the target content profile, as a response to the query request; The determination module 606 is configured to determine the query record information corresponding to the query request and store the query record information in a trusted attribute knowledge base; The update module 608 is configured to update the candidate content profiles in the content profile library based on the query record information in the trusted attribute knowledge base, provided that the content profile library meets the verification conditions.

[0130] In an optional embodiment, the step of constructing a query profile in response to a query request and determining the target content profile associated with the query request in the content profile library based on the query profile includes: Receive query requests and parse them to obtain multi-dimensional query information; Construct a query profile corresponding to the query request based on the multi-dimensional query information; Align the query profile with the candidate content profiles in the content profile library to obtain the target content profile associated with the query request; The candidate content profiles in the content profile library are constructed based on the multi-dimensional content information corresponding to the candidate content.

[0131] In an optional embodiment, aligning the query profile with candidate content profiles in the content profile library to obtain the target content profile associated with the query request includes: Calculate the image similarity between the query image and the candidate content images in the content image library, and filter multiple matching content images in the content image library based on the image similarity. The multiple matching content images are filtered according to the preset image filtering information to obtain a set number of initial content images; The target content profile associated with the query request is determined from the set number of initial content profiles.

[0132] In an optional embodiment, determining the query record information corresponding to the query request and storing the query record information in a trusted attribute knowledge base includes: Determine the query profile corresponding to the query request and the target content profile associated with the query request; Obtain the image attribute structure information and image credibility attribute values ​​corresponding to the query image and the target content image, respectively; Determine the query process information, confidence level, conflict records, and correction information corresponding to the query request; Based on the profile attribute structure information, the profile credibility attribute value, the query process information, the confidence level, the conflict record, and the correction information, construct the query record information corresponding to the query request; The query record information is stored in a trusted attribute knowledge base, wherein the query record information stored in the trusted attribute knowledge base is used to perform retrieval tasks and / or verification tasks.

[0133] In an optional embodiment, updating the candidate content profiles in the content profile library based on query record information in the trusted attribute knowledge base, when the content profile library meets the verification conditions, includes: Based on the query record information stored in the trusted attribute knowledge base, if it is determined that there is a first content profile in the content profile library that has participated more than a set threshold number of times and has not been selected, then the content profile library is determined to meet the verification conditions. The content attribute information to be corrected is determined in the first content profile, and the first query record information associated with the content attribute information to be corrected is read from the trusted attribute knowledge base; Based on the first query record information, determine whether there is an attribute extraction error in the attribute information of the content to be corrected, or determine whether there is an attribute offset in the attribute information of the content to be corrected, or determine whether the attribute information of the content to be corrected does not match the query intent based on the first query record information; If so, adjust the attribute confidence of the content attribute information to be corrected in the first content profile, and add verification mark information for the content attribute information to be corrected; In the offline phase, the evaluation model is used to identify the review marker information, and the content attribute information to be corrected in the first content profile is reviewed and updated based on the recognition results.

[0134] In an optional embodiment, updating the candidate content profiles in the content profile library based on query record information in the trusted attribute knowledge base, when the content profile library meets the verification conditions, includes: Aggregate the candidate content corresponding to the candidate content profile in the content profile library to obtain candidate content clusters; If there are candidate contents with conflicting attributes in the candidate content cluster, it is determined that the content profile library meets the verification conditions; Determine the conflicting content attribute information in the second content profile corresponding to the candidate content with the attribute conflict, and read the second query record information associated with the conflicting content attribute information in the trusted attribute knowledge base; Based on the second query record information, determine whether there are any abnormal items in the conflict content attribute information; If so, adjust the attribute weight of the conflicting content attribute information in the second content profile, and add correction mark information for the conflicting content attribute information; In the offline phase, the evaluation model is used to identify the correction marker information, and the conflict content attribute information in the second content profile is corrected and updated based on the identification results.

[0135] In an optional embodiment, the method further includes: Determine the behavioral information corresponding to the query record information in the trusted attribute knowledge base, and aggregate the behavioral information; If, based on the aggregation results, a set number of behavioral information items are all associated with the same profile attribute information, the weight parameters corresponding to the profile attribute information are adjusted. These weight parameters are used in the query profile construction phase.

[0136] In an optional embodiment, generating query results based on the target content profile as a response to the query request includes: Based on the target content profile, determine the target object associated with the query request from the candidate object set; A list of objects is generated based on the target object, and the list of objects is sent to the user terminal corresponding to the query request as a response to the query request.

[0137] In an optional embodiment, after the steps of determining the query record information corresponding to the query request and storing the query record information in the trusted attribute knowledge base are performed, the method further includes: Time detection is performed on the query record information stored in the trusted attribute knowledge base; Based on the time detection results, multiple query records that exceed the set time threshold are identified as needing optimization. Obtain the retrieval information corresponding to the plurality of query record information to be optimized, and determine the query record information to be deleted from the plurality of query record information to be optimized based on the retrieval information; Delete the query record information to be deleted from the trusted attribute knowledge base.

[0138] The above is an illustrative scheme of an information processing device according to this embodiment. It should be noted that the technical solution of this information processing device and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the information processing device, please refer to the description of the technical solution of the information processing method described above.

[0139] Corresponding to the above method embodiments, this specification also provides another embodiment of an information processing apparatus. Figure 7 A schematic diagram of another information processing apparatus provided in one embodiment of this specification is shown. Figure 7 As shown, the device includes: The profile building module 702 is configured to build an object query profile in response to an object query request, and determine the target object content profile associated with the object query request in the object content profile library based on the object query profile. The list generation module 704 is configured to generate a list of target objects based on the content profile of the target objects, as a response to the object query request; The information determination module 706 is configured to determine the query record information corresponding to the object query request and store the query record information in a trusted attribute knowledge base; The profile update module 708 is configured to update the candidate object content profiles in the object content profile library based on the query record information in the trusted attribute knowledge base, provided that the object content profile library meets the verification conditions.

[0140] The above is an illustrative scheme of another information processing device according to this embodiment. It should be noted that the technical solution of this information processing device and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the information processing device, please refer to the description of the technical solution of the information processing method described above.

[0141] Figure 8 A structural block diagram of a computing device 800 according to one embodiment of this specification is shown. The components of the computing device 800 include, but are not limited to, a memory 810 and a processor 820. The processor 820 is connected to the memory 810 via a bus 830, and a database 850 is used to store data.

[0142] The computing device 800 also includes an access device 840, which enables the computing device 800 to communicate via one or more networks 860. Examples of networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 840 may include one or more of any type of wired or wireless network interface (e.g., a network interface controller (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0143] In one embodiment of this specification, the above-described components of the computing device 800 and Figure 8 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 8The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0144] The computing device 800 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 800 can also be a mobile or stationary server.

[0145] The processor 820 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described information processing method.

[0146] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the information processing method described above.

[0147] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described information processing method.

[0148] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the information processing method described above.

[0149] An embodiment of this specification also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described information processing method.

[0150] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the information processing method described above.

[0151] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or advantageous.

[0152] The computer program / instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0153] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0154] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0155] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. The embodiments selected and specifically described in this specification are intended to better explain the principles and practical applications of these embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An information processing method, comprising: In response to a query request, a query profile is constructed, and based on the query profile, a target content profile associated with the query request is determined in the content profile library. Query results are generated based on the target content profile, serving as a response to the query request; Determine the query record information corresponding to the query request, and store the query record information in the trusted attribute knowledge base; If the content profile database meets the verification conditions, the candidate content profiles in the content profile database are updated based on the query record information in the trusted attribute knowledge base.

2. The information processing method according to claim 1, wherein the step of constructing a query profile in response to a query request and determining a target content profile associated with the query request in a content profile database based on the query profile includes: Receive query requests and parse them to obtain multi-dimensional query information; Construct a query profile corresponding to the query request based on the multi-dimensional query information; Align the query profile with the candidate content profiles in the content profile library to obtain the target content profile associated with the query request; The candidate content profiles in the content profile library are constructed based on the multi-dimensional content information corresponding to the candidate content.

3. The information processing method according to claim 2, wherein aligning the query profile with candidate content profiles in the content profile library to obtain the target content profile associated with the query request includes: Calculate the image similarity between the query image and the candidate content images in the content image library, and filter multiple matching content images in the content image library based on the image similarity. The multiple matching content images are filtered according to the preset image filtering information to obtain a set number of initial content images; The target content profile associated with the query request is determined from the set number of initial content profiles.

4. The information processing method according to claim 1, wherein determining the query record information corresponding to the query request and storing the query record information in a trusted attribute knowledge base includes: Determine the query profile corresponding to the query request and the target content profile associated with the query request; Obtain the image attribute structure information and image credibility attribute values ​​corresponding to the query image and the target content image, respectively; Determine the query process information, confidence level, conflict records, and correction information corresponding to the query request; Based on the profile attribute structure information, the profile credibility attribute value, the query process information, the confidence level, the conflict record, and the correction information, construct the query record information corresponding to the query request; The query record information is stored in a trusted attribute knowledge base, wherein the query record information stored in the trusted attribute knowledge base is used to perform retrieval tasks and / or verification tasks.

5. The information processing method according to claim 1, wherein updating the candidate content profiles in the content profile database based on query record information in the trusted attribute knowledge base when the content profile database meets the verification conditions includes: Based on the query record information stored in the trusted attribute knowledge base, if it is determined that there is a first content profile in the content profile library that has participated more than a set threshold number of times and has not been selected, then the content profile library is determined to meet the verification conditions. The content attribute information to be corrected is determined in the first content profile, and the first query record information associated with the content attribute information to be corrected is read from the trusted attribute knowledge base; Based on the first query record information, determine whether there is an attribute extraction error in the attribute information of the content to be corrected, or determine whether there is an attribute offset in the attribute information of the content to be corrected, or determine whether the attribute information of the content to be corrected does not match the query intent based on the first query record information; If so, adjust the attribute confidence of the content attribute information to be corrected in the first content profile, and add verification mark information for the content attribute information to be corrected; In the offline phase, the evaluation model is used to identify the review marker information, and the content attribute information to be corrected in the first content profile is reviewed and updated based on the recognition results.

6. The information processing method according to claim 1, wherein updating the candidate content profiles in the content profile database based on query record information in the trusted attribute knowledge base when the content profile database meets the verification conditions includes: Aggregate the candidate content corresponding to the candidate content profile in the content profile library to obtain candidate content clusters; If there are candidate contents with conflicting attributes in the candidate content cluster, it is determined that the content profile library meets the verification conditions; Determine the conflicting content attribute information in the second content profile corresponding to the candidate content with the attribute conflict, and read the second query record information associated with the conflicting content attribute information in the trusted attribute knowledge base; Based on the second query record information, determine whether there are any abnormal items in the conflict content attribute information; If so, adjust the attribute weight of the conflicting content attribute information in the second content profile, and add correction mark information for the conflicting content attribute information; In the offline phase, the evaluation model is used to identify the correction marker information, and the conflict content attribute information in the second content profile is corrected and updated based on the identification results.

7. The information processing method according to claim 1, further comprising: Determine the behavioral information corresponding to the query record information in the trusted attribute knowledge base, and aggregate the behavioral information; If, based on the aggregation results, a set number of behavioral information items are all associated with the same profile attribute information, the weight parameters corresponding to the profile attribute information are adjusted. These weight parameters are used in the query profile construction phase.

8. The information processing method according to any one of claims 1 to 7, wherein generating query results based on the target content profile as a response to the query request comprises: Based on the target content profile, determine the target object associated with the query request from the candidate object set; A list of objects is generated based on the target object, and the list of objects is sent to the user terminal corresponding to the query request as a response to the query request.

9. The information processing method according to any one of claims 1 to 7, after the step of determining the query record information corresponding to the query request and storing the query record information in the trusted attribute knowledge base is performed, it further includes: Time detection is performed on the query record information stored in the trusted attribute knowledge base; Based on the time detection results, multiple query records that exceed the set time threshold are identified as needing optimization. Obtain the retrieval information corresponding to the plurality of query record information to be optimized, and determine the query record information to be deleted from the plurality of query record information to be optimized based on the retrieval information; Delete the query record information to be deleted from the trusted attribute knowledge base.

10. An information processing method, comprising: In response to an object query request, an object query profile is constructed, and the target object content profile associated with the object query request is determined in the object content profile library based on the object query profile. A list of target objects is generated based on the content profile of the target objects, which serves as a response to the object query request; Determine the query record information corresponding to the object query request, and store the query record information in a trusted attribute knowledge base; If the object content profile library meets the verification conditions, the candidate object content profiles in the object content profile library are updated based on the query record information in the trusted attribute knowledge base.

11. An information processing system, comprising a user terminal and a server, including: The server is used to construct a query profile in response to a query request submitted by the user terminal, and to determine the target content profile associated with the query request in the content profile library based on the query profile. Query results are generated based on the target content profile, and the query results are fed back to the user terminal; The user terminal is used to display the query results on the management page; The server is also used to determine the query record information corresponding to the query request and store the query record information in a trusted attribute knowledge base; If the content profile database meets the verification conditions, the candidate content profiles in the content profile database are updated based on the query record information in the trusted attribute knowledge base.

12. A computing device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

14. A computer program product comprising a computer program or instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

Images