AI search result analysis method and system based on collaborative feature calculus

Abstract

The application discloses an AI search result analysis method and system based on collaborative feature calculation, which comprises the following steps: obtaining historical media data published in a historical time period and corresponding historical AI question and answer results; based on a model construction algorithm, constructing an AI media analysis model according to the historical media data and the corresponding historical AI question and answer results; in response to a media publishing instruction, sending a search question corresponding to current media publishing content to a target AI platform and obtaining corresponding current AI question and answer results; and analyzing the current AI question and answer results according to the AI media analysis model to obtain a content monitoring result corresponding to the current media publishing content on the target AI platform. It can be seen that the application can realize intelligent monitoring and evaluation of the search question and answer performance of media content on an AI platform, improve the timeliness and accuracy of content risk identification, and provide more accurate data basis for the brand communication effect of the AI platform.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an AI search result analysis method and system based on collaborative feature calculation. Background Technology

[0002] With the deep integration of digital media dissemination and AI Q&A platforms, enterprises and brands are increasingly emphasizing the real-time monitoring and evaluation of media content performance on these platforms through intelligent means. A key technical challenge is how to achieve intelligent identification of the dissemination effects of media content. Existing technologies typically acquire published media data, monitor platform feedback through manual review or simple rule matching, and assess content risk based on fixed templates to support brand communication management. However, existing solutions lack deep model building of historical media data and corresponding AI Q&A results, as well as intelligent correlation analysis of current AI Q&A results. This makes it difficult to achieve dynamic risk assessment of content performance on the platform. Commonly used manual or static monitoring strategies cannot effectively capture the real-time feedback characteristics of AI platforms on media content, resulting in insufficient timeliness and accuracy in content risk identification, thus limiting the effectiveness of intelligent media content monitoring and risk control. Therefore, existing technologies have shortcomings that urgently need to be addressed. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide an AI search result analysis method and system based on collaborative feature calculation, which can realize intelligent monitoring and evaluation of the performance of media content in AI platform search and question answering, improve the timeliness and accuracy of content risk identification, and provide a more accurate data foundation for the brand communication effect of AI platform.

[0004] To address the aforementioned technical problems, the first aspect of this invention discloses an AI search result analysis method based on collaborative feature computation, the method comprising: Obtain historical media data and corresponding historical AI Q&A results published within historical time periods; Based on the model building algorithm, an AI media analysis model is constructed according to the historical media data and the corresponding historical AI question-and-answer results; In response to media release instructions, send the search question corresponding to the current media release content to the target AI platform and obtain the corresponding current AI question-answering results; The AI ​​media analysis model is used to analyze the current AI question-and-answer results to obtain the content monitoring results of the current media content on the target AI platform.

[0005] As an optional implementation, in the first aspect of the present invention, the step of constructing an AI media analysis model based on the model building algorithm, according to the historical media data and the corresponding historical AI question-and-answer results, includes: For each historical media data point, based on a classification model, media release data, brand fact data, and product knowledge data are filtered out from that historical media data. Based on the media release data, brand fact data, and product knowledge data, a corresponding media information matrix is ​​established. The historical media data and its corresponding historical AI question-and-answer results on multiple platforms are input into a trained answer quality analysis model to obtain answer quality information for each historical AI question-and-answer result. The answer quality analysis model is trained using a training dataset that includes multiple training AI question-and-answer results and corresponding answer quality analysis annotations. The answer quality information includes at least one of the following: the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citations, the degree of factual deviation in the answer, and the accuracy of competitor mentions. Using the media information matrix as training data and the corresponding answer quality information as data annotations, a preset basic LLM model is trained to obtain a trained AI media analysis model.

[0006] As an optional implementation, in the first aspect of the invention, the step of filtering media release data, brand fact data, and product knowledge data from the historical media data based on a classification model includes: The historical media data is input into a trained data classification model to obtain multiple media news data and multiple brand promotion data. Based on product-related keywords, multiple product information data are matched from the media news data and the brand promotion data; For each target data, the average text similarity between that target data and all other target data is calculated to obtain the credibility of that target data; When the credibility is greater than a preset credibility threshold, the target data is determined as the corresponding filter data; when the target data is the media news data, the brand promotion data or the product information data, the filter data is respectively the media release data, brand fact data or product knowledge data.

[0007] As an optional implementation, in the first aspect of the present invention, establishing a corresponding media information matrix based on the media release data, brand fact data, and product knowledge data includes: For each product knowledge data, establish the corresponding graph node for that product knowledge data; Calculate the first correlation between each of the media release data and the product knowledge data; Calculate the second correlation between the real-time data of each brand and the product knowledge data; The media release data or the brand real-time data with the first correlation degree or the second correlation degree greater than a preset threshold are determined as the associated data corresponding to the product knowledge data; All the associated data are identified as the node feature data corresponding to the graph nodes; Calculate the data similarity between the node feature data of any two graph nodes to obtain the node association features between the two graph nodes; The graph matrix data composed of all the graph nodes, their corresponding node feature data, and the node association features is determined as the corresponding media information matrix.

[0008] As an optional implementation, in the first aspect of the invention, the search problem is determined through the following steps: Obtain the set of historical questions corresponding to the target AI platform in the historical AI question-and-answer results; Based on the aforementioned set of historical questions, the prompt word rules are determined; Based on the aforementioned prompt word rules, and according to the current media content, the search question is constructed.

[0009] As an optional implementation, in the first aspect of the present invention, the specific step of determining the prompt word rules based on the historical question set includes: Determine the answer quality information for each historical question in the set of historical questions; Based on the answer quality information corresponding to each historical question, and based on the preset scoring rules, determine the answer quality score corresponding to each historical question. Historical questions with answer quality scores greater than a preset score threshold are selected to obtain multiple preferred historical questions; All the aforementioned preferred historical problems are input into the LLM model for problem feature analysis to obtain problem features; Construct suggestion rules that include the characteristics of the aforementioned problem.

[0010] As an optional implementation, in the first aspect of the present invention, the step of analyzing the current AI question-and-answer result based on the AI ​​media analysis model to obtain the content monitoring result corresponding to the current media published content on the target AI platform includes: The current AI question-answering result is input into the AI ​​media analysis model to obtain multiple quality dimension information corresponding to the current AI question-answering result; the quality dimension information includes the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citation, the degree of factual deviation of the answer, or the accuracy of competitor mentions; Filter out the quality dimension information that is lower than the dimension threshold corresponding to the information type to obtain at least one content defect dimension; Input the aforementioned content defect dimensions into the LLM model to obtain the corresponding specific analysis results; Based on the aforementioned content defect dimensions and the corresponding specific analysis results, a content detection report is generated for the current media-published content on the target AI platform.

[0011] As an optional implementation, in the first aspect of the present invention, the method further includes: The content defect dimension and the search question are input into the trained problem improvement model to obtain the output improved question; the problem improvement model is trained using a training dataset that includes multiple training question data and corresponding defect dimension annotations and improved question annotations. The improved question is sent to the target AI platform to obtain new AI question-answering results; The new AI question-and-answer results are analyzed based on the AI ​​media analysis model to obtain the content monitoring update results of the current media content on the target AI platform. Calculate the result difference between the content monitoring update result and the content monitoring result; When the difference in the results is greater than a preset difference threshold, the content monitoring update result is determined as the final monitoring result, and the search question, the improvement question, and the content defect dimension are used as new positive sample data and input into the problem improvement model for optimization.

[0012] A second aspect of this invention discloses an AI search result analysis system based on collaborative feature computation, the system comprising: The first acquisition module is used to acquire historical media data and corresponding historical AI Q&A results published in historical time periods. The model building module is used to build an AI media analysis model based on the model building algorithm, according to the historical media data and the corresponding historical AI question-and-answer results; The second acquisition module is used to respond to media release instructions by sending the search question corresponding to the current media release content to the target AI platform and obtaining the corresponding current AI question answering results; The content analysis module is used to analyze the current AI question-and-answer results based on the AI ​​media analysis model to obtain the content monitoring results of the current media content on the target AI platform.

[0013] As an optional implementation, in the second aspect of the present invention, the specific method by which the model building module constructs an AI media analysis model based on a model building algorithm, according to the historical media data and the corresponding historical AI question-and-answer results, includes: For each historical media data point, based on a classification model, media release data, brand fact data, and product knowledge data are filtered out from that historical media data. Based on the media release data, brand fact data, and product knowledge data, a corresponding media information matrix is ​​established. The historical media data and its corresponding historical AI question-and-answer results on multiple platforms are input into a trained answer quality analysis model to obtain answer quality information for each historical AI question-and-answer result. The answer quality analysis model is trained using a training dataset that includes multiple training AI question-and-answer results and corresponding answer quality analysis annotations. The answer quality information includes at least one of the following: the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citations, the degree of factual deviation in the answer, and the accuracy of competitor mentions. Using the media information matrix as training data and the corresponding answer quality information as data annotations, a preset basic LLM model is trained to obtain a trained AI media analysis model.

[0014] As an optional implementation, in the second aspect of the invention, the specific method by which the model building module filters out media release data, brand fact data, and product knowledge data from the historical media data based on a classification model includes: The historical media data is input into a trained data classification model to obtain multiple media news data and multiple brand promotion data. Based on product-related keywords, multiple product information data are matched from the media news data and the brand promotion data; For each target data, the average text similarity between that target data and all other target data is calculated to obtain the credibility of that target data; When the credibility is greater than a preset credibility threshold, the target data is determined as the corresponding filter data; when the target data is the media news data, the brand promotion data or the product information data, the filter data is respectively the media release data, brand fact data or product knowledge data.

[0015] As an optional implementation, in the second aspect of the present invention, the specific method by which the model building module establishes a corresponding media information matrix based on the media release data, brand fact data, and product knowledge data includes: For each product knowledge data, establish the corresponding graph node for that product knowledge data; Calculate the first correlation between each of the media release data and the product knowledge data; Calculate the second correlation between the real-time data of each brand and the product knowledge data; The media release data or the brand real-time data with the first correlation degree or the second correlation degree greater than a preset threshold are determined as the associated data corresponding to the product knowledge data; All the associated data are identified as the node feature data corresponding to the graph nodes; Calculate the data similarity between the node feature data of any two graph nodes to obtain the node association features between the two graph nodes; The graph matrix data composed of all the graph nodes, their corresponding node feature data, and the node association features is determined as the corresponding media information matrix.

[0016] As an optional implementation, in a second aspect of the invention, the search problem is determined through the following steps: Obtain the set of historical questions corresponding to the target AI platform in the historical AI question-and-answer results; Based on the aforementioned set of historical questions, the prompt word rules are determined; Based on the aforementioned prompt word rules, and according to the current media content, the search question is constructed.

[0017] As an optional implementation, in a second aspect of the invention, the specific step of determining the prompt word rules based on the historical question set includes: Determine the answer quality information for each historical question in the set of historical questions; Based on the answer quality information corresponding to each historical question, and based on the preset scoring rules, determine the answer quality score corresponding to each historical question. Historical questions with answer quality scores greater than a preset score threshold are selected to obtain multiple preferred historical questions; All the aforementioned preferred historical problems are input into the LLM model for problem feature analysis to obtain problem features; Construct suggestion rules that include the characteristics of the aforementioned problem.

[0018] As an optional implementation, in a second aspect of the invention, the specific method by which the content analysis module analyzes the current AI question-and-answer result based on the AI ​​media analysis model to obtain the content monitoring result corresponding to the current media publication content on the target AI platform includes: The current AI question-answering result is input into the AI ​​media analysis model to obtain multiple quality dimension information corresponding to the current AI question-answering result; the quality dimension information includes the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citation, the degree of factual deviation of the answer, or the accuracy of competitor mentions; Filter out the quality dimension information that is lower than the dimension threshold corresponding to the information type to obtain at least one content defect dimension; Input the aforementioned content defect dimensions into the LLM model to obtain the corresponding specific analysis results; Based on the aforementioned content defect dimensions and the corresponding specific analysis results, a content detection report is generated for the current media-published content on the target AI platform.

[0019] As an optional implementation, in a second aspect of the invention, the system is further configured to perform the following steps: The content defect dimension and the search question are input into the trained problem improvement model to obtain the output improved question; the problem improvement model is trained using a training dataset that includes multiple training question data and corresponding defect dimension annotations and improved question annotations. The improved question is sent to the target AI platform to obtain new AI question-answering results; The new AI question-and-answer results are analyzed based on the AI ​​media analysis model to obtain the content monitoring update results of the current media content on the target AI platform. Calculate the result difference between the content monitoring update result and the content monitoring result; When the difference in the results is greater than a preset difference threshold, the content monitoring update result is determined as the final monitoring result, and the search question, the improvement question, and the content defect dimension are used as new positive sample data and input into the problem improvement model for optimization.

[0020] A third aspect of this invention discloses another AI search result analysis system based on collaborative feature computation, the system comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in the AI ​​search result analysis method based on collaborative feature computation disclosed in the first aspect of the present invention.

[0021] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the AI ​​search result analysis method based on collaborative feature computation disclosed in the first aspect of the present invention.

[0022] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention acquires historical media data and corresponding historical AI Q&A results for different time periods, constructs an AI media analysis model based on a model building algorithm, responds to media release instructions by sending search questions to the target AI platform and obtaining the current AI Q&A results, and analyzes the current AI Q&A results based on the model to obtain content monitoring results. This enables intelligent monitoring and evaluation of media content performance in AI platform search Q&A, improves the timeliness and accuracy of content risk identification, and provides a more accurate data foundation for the brand communication effect of AI platforms. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating an AI search result analysis method based on collaborative feature computation disclosed in an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the structure of an AI search result analysis system based on collaborative feature computation disclosed in an embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of another AI search result analysis system based on collaborative feature computation disclosed in an embodiment of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0029] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0030] This invention discloses an AI search result analysis method and system based on collaborative feature computation. By acquiring historical media data and corresponding historical AI question-and-answer results for historical time periods, an AI media analysis model is constructed based on a model building algorithm. Responding to media release instructions, a search question is sent to the target AI platform, and the current AI question-and-answer results are obtained. The model is then used to analyze the current AI question-and-answer results to obtain content monitoring results. This enables intelligent monitoring and evaluation of media content performance in AI platform search questions and answers, improving the timeliness and accuracy of content risk identification and providing a more accurate data foundation for the brand communication effectiveness of AI platforms. Detailed explanations follow.

[0031] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating an AI search result analysis method based on collaborative feature computation disclosed in an embodiment of the present invention. Wherein, Figure 1 The described AI search result analysis method based on collaborative feature calculus can be applied to data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 1 As shown, this AI search result analysis method based on collaborative feature computation can include the following operations: 101. Obtain historical media data and corresponding historical AI Q&A results published within a historical time period.

[0032] Optionally, the historical media data can be IT technology media press releases published across the entire network in the past 180 days, official public relations press releases of specific car brands, KOL review articles and images on social media, or review reports from vertical industry forums. This invention does not limit the scope of the data.

[0033] Optionally, the historical AI question-and-answer results can be native text snapshots generated periodically by requesting mainstream large model retrieval systems via API, or RAG-generated paragraphs with related citations extracted by the crawler engine from intelligent search results. This invention does not limit the scope of these paragraphs.

[0034] 102. Based on the model building algorithm, construct an AI media analysis model according to historical media data and corresponding historical AI question-and-answer results.

[0035] 103. In response to media release instructions, send the search question corresponding to the current media release content to the target AI platform and obtain the corresponding current AI question-and-answer results.

[0036] 104. Analyze the current AI question-and-answer results based on the AI ​​media analysis model to obtain the content monitoring results of the current media content on the target AI platform.

[0037] As can be seen, the above-described embodiments of the invention acquire historical media data and corresponding historical AI Q&A results published in historical time periods, construct an AI media analysis model based on a model building algorithm, respond to media release instructions by sending search questions to the target AI platform and obtaining current AI Q&A results, and analyze the current AI Q&A results according to the model to obtain content monitoring results. This enables intelligent monitoring and evaluation of media content performance in AI platform search Q&A, improves the timeliness and accuracy of content risk identification, and provides a more accurate data foundation for the brand communication effect of the AI ​​platform.

[0038] As an optional embodiment, the above steps, including constructing an AI media analysis model based on the model building algorithm and historical media data and corresponding historical AI question-and-answer results, include: For each historical media data, based on the classification model, media release data, brand fact data, and product knowledge data are filtered out from that historical media data; Establish a corresponding media information matrix based on media release data, brand factual data, and product knowledge data; The historical AI question-and-answer results corresponding to this historical media data on multiple platforms are input into a trained answer quality analysis model to obtain the answer quality information corresponding to each historical AI question-and-answer result; Using the media information matrix as training data and the corresponding answer quality information as the corresponding data annotation, a pre-set basic LLM model is trained to obtain a trained AI media analysis model.

[0039] Optionally, the answer quality analysis model is trained using a training dataset that includes multiple training AI question-answering results and corresponding answer quality analysis annotations.

[0040] Optionally, answer quality information may include at least one of the following: the percentage of text mentioning the brand, the percentage of text presenting product information, the accuracy of media source citations, the degree of factual deviation in the answer, and the accuracy of competitor mentions.

[0041] Optionally, the underlying LLM model can be the open-source Llama-3-8B-Instruct base model, the Qwen2.5-14B-Dense dialogue model, or the Mistral large language model with instruction fine-tuning (SFT). This invention does not limit the specific model.

[0042] Optionally, the training of the preset base LLM model can be performed using a parameter-efficient fine-tuning technique based on low-rank adaptation (LoRA), a fully supervised fine-tuning technique, or an alignment training method based on direct preference optimization (DPO). This invention does not limit the method.

[0043] As can be seen, through the above optional embodiments, by filtering media release data, brand fact data, and product knowledge data based on a classification model for each historical media data, a media information matrix is ​​established. The historical AI question-and-answer results from multiple platforms are input into the answer quality analysis model to obtain answer quality information. The matrix and quality information are used as training data to optimize the basic LLM model to obtain the AI ​​media analysis model. This achieves structured correlation modeling between historical media and question-and-answer results, improves the model's multi-dimensional learning ability for content quality, and enhances the reliability and adaptability of subsequent monitoring and analysis.

[0044] As an optional embodiment, the above steps, based on a classification model, filter out media release data, brand factual data, and product knowledge data from the historical media data, including: The historical media data is input into a trained data classification model to obtain multiple media news data and multiple brand promotion data. Based on product-related keywords, multiple product information data are matched from media news data and brand promotion data; For each target data, calculate the average text similarity between that target data and all other target data to obtain the credibility of that target data; When the credibility is greater than the preset credibility threshold, the target data is determined as the corresponding filter data; when the target data are media news data, brand promotion data or product information data, the filter data are respectively media release data, brand fact data or product knowledge data.

[0045] Optionally, the data classification model can be a multi-label text classification network based on the DeBERTa-v3 architecture, or a BERT semantic slice classifier optimized for long texts; this invention does not impose any limitations.

[0046] Optionally, the preset confidence threshold can be a dynamic confidence boundary set between 0.78 and 0.88 to eliminate low-quality web crawler plagiarism texts and false spam information. This invention does not limit this.

[0047] As can be seen, through the above optional embodiments, by inputting historical media data into the data classification model to obtain media news data and brand promotion data, matching product information data based on product keywords, calculating the average text similarity of target data with other data to obtain credibility, and determining the data to be filtered when the credibility is greater than the threshold, the credibility of media content can be quantitatively filtered, improving the accuracy and automation level of data quality control, and strengthening the reliability and effectiveness of subsequent analysis results.

[0048] As an optional embodiment, the step above, establishing a corresponding media information matrix based on media release data, brand factual data, and product knowledge data, includes: For each product knowledge data, create the corresponding graph node for that product knowledge data; Calculate the first correlation between each media release data and the product knowledge data; Calculate the second degree of correlation between real-time data for each brand and the product knowledge data; Media release data or brand real-time data with a first or second relevance greater than a preset threshold are identified as the associated data corresponding to the product knowledge data. All related data are identified as the node feature data corresponding to the graph nodes; Calculate the data similarity between the node feature data of any two graph nodes to obtain the node association features between the two graph nodes; The graph matrix data consisting of all graph nodes, their corresponding node feature data, and node association features is determined as the corresponding media information matrix.

[0049] Optionally, the graph node can be an entity vector node in a graph neural network (GNN) that represents the core selling points, hardware parameters, or proprietary technical terms of a specific product; this invention does not impose any limitations on this.

[0050] Optionally, the first and second relevance scores can be calculated by reciprocal of the co-occurrence frequency of named entities in two text blocks, or by context relevance scores calculated using a pre-trained cross-encoder. This invention does not limit the specific relevance scores.

[0051] Optionally, the graph matrix data can be represented as a composite tensor data structure composed of a graph adjacency matrix and a high-dimensional node feature matrix, which is not limited in this invention.

[0052] As can be seen, through the above optional embodiments, by establishing graph nodes for each product knowledge data, calculating its first and second correlation degrees with media release data and brand real-time data, determining the associated data, and establishing node feature data and node association features, a graph matrix data is formed as a media information matrix, realizing graph-structured association modeling of media knowledge, improving the expressive ability and analytical depth of data dependencies, and strengthening the relevance and comprehensiveness of content monitoring.

[0053] As an optional embodiment, the search problem in the above steps is determined through the following steps: Obtain the set of historical questions corresponding to the target AI platform in its historical AI question-answering results; Based on the historical question set, the rules for identifying prompt words were determined; Based on the prompt word rules, a search question is constructed according to the current media content.

[0054] Optionally, the prompt rule can be a structured prompt generation framework, such as a meta-prompt template that defines the role setting, context injection slot, and forced output format (JSON / Markdown). This invention does not impose any limitations on this framework.

[0055] As can be seen, through the above optional embodiments, by obtaining the historical question set corresponding to the historical AI question-and-answer results of the target AI platform, determining the prompt word rules based on the set, and constructing search questions based on the rules and the current media content, targeted search questions can be intelligently generated, thereby improving the accuracy and relevance of the platform's question-and-answer acquisition and enhancing the targeting and efficiency of subsequent content monitoring.

[0056] As an optional embodiment, the specific step of determining the prompt word rules based on the historical question set in the above steps includes: Determine the quality information of the answers for each historical question in the set of historical questions; Based on the answer quality information corresponding to each historical question, and based on the preset scoring rules, determine the answer quality score corresponding to each historical question. Historical questions with answer quality scores greater than a preset score threshold are selected to obtain several preferred historical questions; All preferred historical problems are input into the LLM model for problem feature analysis to obtain problem characteristics; Construct rules for prompt words that include problematic features.

[0057] Optionally, the preset scoring rule can be a comprehensive weighted linear summation logic that assigns 40% weight to brand mention rate, 40% weight to factual accuracy, and 20% weight to source citation standardization. This invention does not limit this.

[0058] Optionally, the question features can be the grammatical topology of highly expressive question sentences, the core set of guiding words that triggers deep RAG retrieval of large models, or industry-specific follow-up question logic operators; this invention does not limit these features.

[0059] As can be seen, through the above optional embodiments, by determining the answer quality information corresponding to each historical question in the historical question set, determining the answer quality score based on the scoring rules, selecting and selecting the best historical questions to input into the LLM model for question feature analysis, and constructing prompt word rules containing question features, the feature extraction and prompt word optimization of high-quality historical questions are realized, thereby improving the quality and applicability of search questions and enhancing the intelligent level of content monitoring.

[0060] As an optional embodiment, the step above, analyzing the current AI question-and-answer results based on the AI ​​media analysis model to obtain the content monitoring results of the current media content on the target AI platform, includes: Input the current AI question-and-answer results into the AI ​​media analysis model to obtain multiple quality dimensions of the output current AI question-and-answer results; optionally, the quality dimensions are the proportion of brand mention text, the proportion of product information text, the accuracy of media source citation, the degree of factual deviation of the answer, or the accuracy of competitor mentions; Filter out quality dimension information that is below the dimension threshold corresponding to the information type to obtain at least one content defect dimension. Input the content defect dimension into the LLM model to obtain the corresponding specific analysis results; Based on the dimensions of content defects and the corresponding specific analysis results, a content detection report is generated for the current media content on the target AI platform.

[0061] Optionally, the content defect dimension can be determined by judging whether the content generated by the AI ​​platform contains "serious reverse misleading by competitors", "missing reference to the core technology source of the brand", or "basic factual illusion". This invention does not limit this.

[0062] Optionally, the specific analysis result can be a deep semantic diagnosis output by the LLM model, such as pointing out that "due to the existence of a large number of plagiarized articles containing outdated parameters on the Internet, the target AI platform cited the wrong product price when answering," which is not limited in this invention.

[0063] As can be seen, through the above optional embodiments, by inputting the current AI question-and-answer results into the AI ​​media analysis model to obtain multiple quality dimension information, filtering out content defect dimensions below the dimension threshold and inputting them into the LLM model to obtain specific analysis results, and generating a content detection report based on the defect dimensions and analysis results, the platform's question-and-answer results can be evaluated for multiple dimensions and analyzed for defects, thereby improving the granularity and interpretability of content monitoring and enhancing the timeliness of risk identification and the effectiveness of decision support.

[0064] As an optional embodiment, the method further includes the following steps: The content defect dimension and search question are input into the trained problem improvement model to obtain the output improved question; optionally, the problem improvement model is trained using a training dataset that includes multiple training question data and corresponding defect dimension annotations and improvement question annotations. Send the improvement question to the target AI platform to obtain new AI question-and-answer results; The new AI question-and-answer results are analyzed based on the AI ​​media analysis model to obtain the content monitoring and update results of the current media content on the target AI platform. Calculate the difference between the content monitoring update results and the content monitoring results; When the difference in results exceeds the preset difference threshold, the content monitoring update result is determined as the final monitoring result, and the search issues, improvement issues, and content defect dimensions are used as new positive sample data and input into the issue improvement model for optimization.

[0065] Optionally, the improved model for this problem can be a Prompt rewrite model based on deep fine-tuning of the T5 architecture, or a question bias corrector trained using the reinforcement learning proximal policy optimization (PPO) algorithm; this invention does not limit the specific model.

[0066] Optionally, the difference in results can be the Jansen-Shannon divergence (JS Divergence) of the two generated versions on the score radar charts, or the absolute percentage increase in brand SOV (sound share), which is not limited in this invention.

[0067] Optionally, the new positive sample data can be represented as a high-quality aligned triplet feature stream in the form of <original search problem, detected content defects, optimized improvement problem>, which can be used to trigger online fine-tuning iterations of the improvement model for the problem. This invention does not limit this.

[0068] As can be seen, through the above optional embodiments, by inputting the content defect dimension and search question into the question improvement model to obtain the improved question, sending it to the platform to obtain new AI question answering results and analyzing them to obtain monitoring update results, calculating the difference between the updated results and the original results, and determining the final monitoring result when the difference is greater than a threshold, and using the data as positive samples to optimize the model, iterative content monitoring and model self-optimization are achieved, improving the accuracy and adaptability of monitoring, and strengthening the continuity and effectiveness of the platform's content risk prevention and control.

[0069] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of an AI search result analysis system based on collaborative feature computation disclosed in an embodiment of the present invention. Figure 2 The described AI search result analysis system based on collaborative feature computation can be applied to data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 2 As shown, this AI search result analysis system based on collaborative feature computation may include: The first acquisition module 201 is used to acquire historical media data published in historical time periods and corresponding historical AI question-and-answer results.

[0070] The model building module 202 is used to build an AI media analysis model based on the model building algorithm, historical media data and corresponding historical AI question-and-answer results.

[0071] The second acquisition module 203 is used to respond to the media release instruction, send the search question corresponding to the current media release content to the target AI platform, and obtain the corresponding current AI question answering result.

[0072] The content analysis module 204 is used to analyze the current AI question-and-answer results based on the AI ​​media analysis model in order to obtain the content monitoring results of the current media content on the target AI platform.

[0073] As can be seen, the above-described embodiments of the invention acquire historical media data and corresponding historical AI Q&A results published in historical time periods, construct an AI media analysis model based on a model building algorithm, respond to media release instructions by sending search questions to the target AI platform and obtaining current AI Q&A results, and analyze the current AI Q&A results according to the model to obtain content monitoring results. This enables intelligent monitoring and evaluation of media content performance in AI platform search Q&A, improves the timeliness and accuracy of content risk identification, and provides a more accurate data foundation for the brand communication effect of the AI ​​platform.

[0074] As an optional implementation, the model building module, based on the model building algorithm, constructs the AI ​​media analysis model in the following specific ways according to historical media data and corresponding historical AI question-and-answer results: For each historical media data, based on the classification model, media release data, brand fact data, and product knowledge data are filtered out from that historical media data; Establish a corresponding media information matrix based on media release data, brand factual data, and product knowledge data; The historical media data and the corresponding historical AI question-and-answer results on multiple platforms are input into a trained answer quality analysis model to obtain answer quality information for each historical AI question-and-answer result. Optionally, the answer quality analysis model is trained using a training dataset that includes multiple training AI question-and-answer results and corresponding answer quality analysis annotations. The answer quality information includes at least one of the following: the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citation, the degree of factual deviation in the answer, and the accuracy of competitor mentions. Using the media information matrix as training data and the corresponding answer quality information as the corresponding data annotation, a pre-set basic LLM model is trained to obtain a trained AI media analysis model.

[0075] As can be seen, through the above optional embodiments, by filtering media release data, brand fact data, and product knowledge data based on a classification model for each historical media data, a media information matrix is ​​established. The historical AI question-and-answer results from multiple platforms are input into the answer quality analysis model to obtain answer quality information. The matrix and quality information are used as training data to optimize the basic LLM model to obtain the AI ​​media analysis model. This achieves structured correlation modeling between historical media and question-and-answer results, improves the model's multi-dimensional learning ability for content quality, and enhances the reliability and adaptability of subsequent monitoring and analysis.

[0076] As an optional implementation, the model building module, based on a classification model, filters out media release data, brand fact data, and product knowledge data from the historical media data in the following specific ways: The historical media data is input into a trained data classification model to obtain multiple media news data and multiple brand promotion data. Based on product-related keywords, multiple product information data are matched from media news data and brand promotion data; For each target data, calculate the average text similarity between that target data and all other target data to obtain the credibility of that target data; When the credibility is greater than the preset credibility threshold, the target data is determined as the corresponding filter data; when the target data are media news data, brand promotion data or product information data, the filter data are respectively media release data, brand fact data or product knowledge data.

[0077] As can be seen, through the above optional embodiments, by inputting historical media data into the data classification model to obtain media news data and brand promotion data, matching product information data based on product keywords, calculating the average text similarity of target data with other data to obtain credibility, and determining the data to be filtered when the credibility is greater than the threshold, the credibility of media content can be quantitatively filtered, improving the accuracy and automation level of data quality control, and strengthening the reliability and effectiveness of subsequent analysis results.

[0078] As an optional implementation, the specific methods by which the model building module establishes the corresponding media information matrix based on media release data, brand factual data, and product knowledge data include: For each product knowledge data, create the corresponding graph node for that product knowledge data; Calculate the first correlation between each media release data and the product knowledge data; Calculate the second degree of correlation between real-time data for each brand and the product knowledge data; Media release data or brand real-time data with a first or second relevance greater than a preset threshold are identified as the associated data corresponding to the product knowledge data. All related data are identified as the node feature data corresponding to the graph nodes; Calculate the data similarity between the node feature data of any two graph nodes to obtain the node association features between the two graph nodes; The graph matrix data consisting of all graph nodes, their corresponding node feature data, and node association features is determined as the corresponding media information matrix.

[0079] As can be seen, through the above optional embodiments, by establishing graph nodes for each product knowledge data, calculating its first and second correlation degrees with media release data and brand real-time data, determining the associated data, and establishing node feature data and node association features, a graph matrix data is formed as a media information matrix, realizing graph-structured association modeling of media knowledge, improving the expressive ability and analytical depth of data dependencies, and strengthening the relevance and comprehensiveness of content monitoring.

[0080] As an optional embodiment, the search problem is determined through the following steps: Obtain the set of historical questions corresponding to the target AI platform in its historical AI question-answering results; Based on the historical question set, the rules for identifying prompt words were determined; Based on the prompt word rules, a search question is constructed according to the current media content.

[0081] As can be seen, through the above optional embodiments, by obtaining the historical question set corresponding to the historical AI question-and-answer results of the target AI platform, determining the prompt word rules based on the set, and constructing search questions based on the rules and the current media content, targeted search questions can be intelligently generated, thereby improving the accuracy and relevance of the platform's question-and-answer acquisition and enhancing the targeting and efficiency of subsequent content monitoring.

[0082] As an optional implementation, the specific steps for determining the prompt word rules based on the historical question set include: Determine the quality information of the answers for each historical question in the set of historical questions; Based on the answer quality information corresponding to each historical question, and based on the preset scoring rules, determine the answer quality score corresponding to each historical question. Historical questions with answer quality scores greater than a preset score threshold are selected to obtain several preferred historical questions; All preferred historical problems are input into the LLM model for problem feature analysis to obtain problem characteristics; Construct rules for prompt words that include problematic features.

[0083] As can be seen, through the above optional embodiments, by determining the answer quality information corresponding to each historical question in the historical question set, determining the answer quality score based on the scoring rules, selecting and selecting the best historical questions to input into the LLM model for question feature analysis, and constructing prompt word rules containing question features, the feature extraction and prompt word optimization of high-quality historical questions are realized, thereby improving the quality and applicability of search questions and enhancing the intelligent level of content monitoring.

[0084] As an optional embodiment, the content analysis module analyzes the current AI question-and-answer results based on the AI ​​media analysis model to obtain the specific method of the content monitoring results of the current media content on the target AI platform, including: Input the current AI question-and-answer results into the AI ​​media analysis model to obtain multiple quality dimensions of the output current AI question-and-answer results; optionally, the quality dimensions are the proportion of brand mention text, the proportion of product information text, the accuracy of media source citation, the degree of factual deviation of the answer, or the accuracy of competitor mentions; Filter out quality dimension information that is below the dimension threshold corresponding to the information type to obtain at least one content defect dimension. Input the content defect dimension into the LLM model to obtain the corresponding specific analysis results; Based on the dimensions of content defects and the corresponding specific analysis results, a content detection report is generated for the current media content on the target AI platform.

[0085] As can be seen, through the above optional embodiments, by inputting the current AI question-and-answer results into the AI ​​media analysis model to obtain multiple quality dimension information, filtering out content defect dimensions below the dimension threshold and inputting them into the LLM model to obtain specific analysis results, and generating a content detection report based on the defect dimensions and analysis results, the platform's question-and-answer results can be evaluated for multiple dimensions and analyzed for defects, thereby improving the granularity and interpretability of content monitoring and enhancing the timeliness of risk identification and the effectiveness of decision support.

[0086] As an optional embodiment, the system is also used to perform the following steps: The content defect dimension and search question are input into the trained problem improvement model to obtain the output improved question; optionally, the problem improvement model is trained using a training dataset that includes multiple training question data and corresponding defect dimension annotations and improvement question annotations. Send the improvement question to the target AI platform to obtain new AI question-and-answer results; The new AI question-and-answer results are analyzed based on the AI ​​media analysis model to obtain the content monitoring and update results of the current media content on the target AI platform. Calculate the difference between the content monitoring update results and the content monitoring results; When the difference in results exceeds the preset difference threshold, the content monitoring update result is determined as the final monitoring result, and the search issues, improvement issues, and content defect dimensions are used as new positive sample data and input into the issue improvement model for optimization.

[0087] As can be seen, through the above optional embodiments, by inputting the content defect dimension and search question into the question improvement model to obtain the improved question, sending it to the platform to obtain new AI question answering results and analyzing them to obtain monitoring update results, calculating the difference between the updated results and the original results, and determining the final monitoring result when the difference is greater than a threshold, and using the data as positive samples to optimize the model, iterative content monitoring and model self-optimization are achieved, improving the accuracy and adaptability of monitoring, and strengthening the continuity and effectiveness of the platform's content risk prevention and control.

[0088] Example 3 Please see Figure 3 , Figure 3 This is another AI search result analysis system based on collaborative feature computation disclosed in the embodiments of the present invention. Figure 3 The described AI search result analysis system based on collaborative feature computation is applied in data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 3 As shown, this AI search result analysis system based on collaborative feature computation may include: Memory 301 storing executable program code; Processor 302 coupled to memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute the steps of the AI ​​search result analysis method based on collaborative feature calculation described in Embodiment 1.

[0089] Example 4 This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps of the AI ​​search result analysis method based on collaborative feature computation described in Embodiment 1.

[0090] Example 5 This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps of the AI ​​search result analysis method based on collaborative feature computation described in Embodiment 1.

[0091] 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 described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily have to follow the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0092] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0093] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0094] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0096] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0098] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0099] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0100] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0101] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0102] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0103] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0104] Finally, it should be noted that the AI ​​search result analysis method and system based on collaborative feature computation disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI search result analysis method based on collaborative feature computation, characterized in that, The method includes: Obtain historical media data and corresponding historical AI Q&A results published within historical time periods; Based on the model building algorithm, an AI media analysis model is constructed according to the historical media data and the corresponding historical AI question-and-answer results; In response to media release instructions, send the search question corresponding to the current media release content to the target AI platform and obtain the corresponding current AI question-answering results; The AI ​​media analysis model is used to analyze the current AI question-and-answer results to obtain the content monitoring results of the current media content on the target AI platform.

2. The AI ​​search result analysis method based on collaborative feature computation according to claim 1, characterized in that, The model-based algorithm constructs an AI media analysis model based on the historical media data and corresponding historical AI question-and-answer results, including: For each historical media data point, based on a classification model, media release data, brand fact data, and product knowledge data are filtered out from that historical media data. Based on the media release data, brand fact data, and product knowledge data, a corresponding media information matrix is ​​established. The historical media data and its corresponding historical AI question-and-answer results on multiple platforms are input into a trained answer quality analysis model to obtain answer quality information for each historical AI question-and-answer result. The answer quality analysis model is trained using a training dataset that includes multiple training AI question-and-answer results and corresponding answer quality analysis annotations. The answer quality information includes at least one of the following: the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citations, the degree of factual deviation in the answer, and the accuracy of competitor mentions. Using the media information matrix as training data and the corresponding answer quality information as data annotations, a preset basic LLM model is trained to obtain a trained AI media analysis model.

3. The AI ​​search result analysis method based on collaborative feature computation according to claim 2, characterized in that, The classification model is used to filter out media release data, brand factual data, and product knowledge data from the historical media data, including: The historical media data is input into a trained data classification model to obtain multiple media news data and multiple brand promotion data. Based on product-related keywords, multiple product information data are matched from the media news data and the brand promotion data; For each target data, the average text similarity between that target data and all other target data is calculated to obtain the credibility of that target data; When the credibility is greater than a preset credibility threshold, the target data is determined as the corresponding filter data; when the target data is the media news data, the brand promotion data or the product information data, the filter data is respectively the media release data, brand fact data or product knowledge data.

4. The AI ​​search result analysis method based on collaborative feature computation according to claim 2, characterized in that, The step of establishing a corresponding media information matrix based on the media release data, brand fact data, and product knowledge data includes: For each product knowledge data, establish the corresponding graph node for that product knowledge data; Calculate the first correlation between each of the media release data and the product knowledge data; Calculate the second correlation between the real-time data of each brand and the product knowledge data; The media release data or the brand real-time data with the first correlation degree or the second correlation degree greater than a preset threshold are determined as the associated data corresponding to the product knowledge data; All the associated data are identified as the node feature data corresponding to the graph nodes; Calculate the data similarity between the node feature data of any two graph nodes to obtain the node association features between the two graph nodes; The graph matrix data composed of all the graph nodes, their corresponding node feature data, and the node association features is determined as the corresponding media information matrix.

5. The AI ​​search result analysis method based on collaborative feature computation according to claim 1, characterized in that, The search question is determined through the following steps: Obtain the set of historical questions corresponding to the target AI platform in the historical AI question-and-answer results; Based on the aforementioned set of historical questions, the prompt word rules are determined; Based on the aforementioned prompt word rules, and according to the current media content, the search question is constructed.

6. The AI ​​search result analysis method based on collaborative feature computation according to claim 5, characterized in that, The specific steps for determining the prompt word rules based on the historical question set include: Determine the answer quality information for each historical question in the set of historical questions; Based on the answer quality information corresponding to each historical question, and based on the preset scoring rules, determine the answer quality score corresponding to each historical question. Historical questions with answer quality scores greater than a preset score threshold are selected to obtain multiple preferred historical questions; All the aforementioned preferred historical problems are input into the LLM model for problem feature analysis to obtain problem features; Construct suggestion rules that include the characteristics of the aforementioned problem.

7. The AI ​​search result analysis method based on collaborative feature computation according to claim 1, characterized in that, The step of analyzing the current AI question-and-answer results based on the AI ​​media analysis model to obtain the content monitoring results of the current media content on the target AI platform includes: The current AI question-answering result is input into the AI ​​media analysis model to obtain multiple quality dimension information corresponding to the current AI question-answering result; the quality dimension information includes the proportion of brand mention text, the proportion of product information presentation text, the accuracy of media source citation, the degree of factual deviation of the answer, or the accuracy of competitor mentions; Filter out the quality dimension information that is lower than the dimension threshold corresponding to the information type to obtain at least one content defect dimension; Input the aforementioned content defect dimensions into the LLM model to obtain the corresponding specific analysis results; Based on the aforementioned content defect dimensions and the corresponding specific analysis results, a content detection report is generated for the current media-published content on the target AI platform.

8. The AI ​​search result analysis method based on collaborative feature computation according to claim 7, characterized in that, The method further includes: The content defect dimension and the search question are input into the trained problem improvement model to obtain the output improved question; the problem improvement model is trained using a training dataset that includes multiple training question data and corresponding defect dimension annotations and improved question annotations. The improved question is sent to the target AI platform to obtain new AI question-answering results; The new AI question-and-answer results are analyzed based on the AI ​​media analysis model to obtain the content monitoring update results of the current media content on the target AI platform. Calculate the result difference between the content monitoring update result and the content monitoring result; When the difference in the results is greater than a preset difference threshold, the content monitoring update result is determined as the final monitoring result, and the search question, the improvement question, and the content defect dimension are used as new positive sample data and input into the problem improvement model for optimization.

9. An AI search result analysis system based on collaborative feature computation, characterized in that, The system includes: The first acquisition module is used to acquire historical media data and corresponding historical AI Q&A results published in historical time periods. The model building module is used to build an AI media analysis model based on the model building algorithm, according to the historical media data and the corresponding historical AI question-and-answer results; The second acquisition module is used to respond to media release instructions by sending the search question corresponding to the current media release content to the target AI platform and obtaining the corresponding current AI question answering results; The content analysis module is used to analyze the current AI question-and-answer results based on the AI ​​media analysis model to obtain the content monitoring results of the current media content on the target AI platform.

10. An AI search result analysis system based on collaborative feature calculus, characterized in that, The system includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the AI ​​search result analysis method based on collaborative feature computation as described in any one of claims 1-8.

Images