A mobile phone product recommendation method based on comparison of question and answer text and comparison of video

Abstract

The application relates to a mobile phone product recommendation method based on comparison of question and answer text and comparison of video, a first comprehensive intuitionistic fuzzy number of a comparison question corresponding to an attribute is obtained based on a first emotion scoring agent; a mean value of the first comprehensive intuitionistic fuzzy numbers of all comparison questions corresponding to the attribute is calculated to obtain a first fuzzy number of a mobile phone product under the corresponding attribute; complete video content of an evaluation comparison video corresponding to the mobile phone product is obtained, and video text information of the complete video content is extracted; the video text information is split into multiple text parts according to the attribute, a second comprehensive intuitionistic fuzzy number of a text part corresponding to the attribute is obtained based on a second emotion scoring agent; the second comprehensive intuitionistic fuzzy numbers of all text parts corresponding to the attribute are aggregated to obtain a second fuzzy number of the mobile phone product under the corresponding attribute; a recommendation score of the mobile phone product is calculated based on the first fuzzy numbers and the second fuzzy numbers under all attributes, and a mobile phone product with the highest recommendation score is taken as a recommended product.

Description

Technical Field

[0001] This application relates to the field of mobile phone product recommendation technology, and in particular to a mobile phone product recommendation method based on comparative question-and-answer text and comparative video. Background Technology

[0002] In today's rapidly developing mobile internet and increasingly competitive smartphone market, consumers often face a dilemma when confronted with a dazzling array of models and differentiated configurations. Online community Q&A platforms address this by having potential consumers ask questions about core metrics such as phone performance, battery life, and camera capabilities. Existing users and other interested parties provide answers based on their real-world experiences and online information, offering extensive comparisons between target models and competitors. This type of community Q&A information, derived from real-world consumer scenarios, provides more authentic and comprehensive content, offering direct reference suggestions for potential buyers. Furthermore, with the rapid development of internet technology, short video platforms have experienced explosive growth. Numerous video bloggers publish review and comparison videos, visually showcasing the differences in performance across various dimensions of different phones, clearly outlining the strengths and weaknesses of each model, and providing more professional and systematic decision-making references for users with purchasing needs.

[0003] However, existing recommendation methods have two limitations: firstly, they fail to organically integrate real consumer feedback from community Q&A sessions with the objective presentation of review videos; secondly, they lack effective methods to transform this information into a basis for recommendations, forcing consumers to browse and analyze fragmented information on their own; and faced with a massive amount of information, consumers find it difficult to quickly obtain comprehensive and valuable mobile phone purchasing advice. Summary of the Invention

[0004] Therefore, it is necessary to provide a mobile product recommendation method based on comparative question-and-answer text and comparative videos, including:

[0005] S1: Obtain product comparison Q&A text information from user community Q&A on online platforms, and construct comparison Q&A text containing comparison questions and their answers;

[0006] S2: Classify all comparison question and answer texts according to the attributes of the mobile phone product, obtain the first comprehensive intuition fuzzy number of the comparison questions corresponding to the attributes based on the first sentiment score agent; calculate the mean of the first comprehensive intuition fuzzy number of all comparison questions corresponding to the attributes to obtain the first fuzzy number of the mobile phone product under the corresponding attributes;

[0007] S3: Obtain the complete video content of the evaluation and comparison video of the corresponding mobile phone product, and extract the video text information of the complete video content;

[0008] S4: Split the video text information into multiple text parts according to the attributes, and obtain the second comprehensive intuition fuzzy number of the text part corresponding to the attribute based on the second sentiment scoring agent; aggregate the second comprehensive intuition fuzzy numbers of all text parts corresponding to the attribute to obtain the second fuzzy number of the mobile phone product under the corresponding attribute.

[0009] S5: Based on the first and second fuzzy numbers under all attributes, calculate the recommendation score of the mobile phone product, and select the mobile phone product with the highest recommendation score as the recommended product.

[0010] Beneficial effects: By integrating community Q&A and comparison videos, the accuracy and comprehensiveness of recommendations are improved, providing users with more suitable and valuable mobile phone products that better meet their needs, helping them to accurately select the right product from among many mobile phone models. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a mobile product recommendation method based on comparing question-and-answer text and comparing videos, as described in this application. Detailed Implementation

[0013] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0014] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0015] like Figure 1 As shown, this embodiment provides a mobile product recommendation method based on comparing question-and-answer text and comparing videos, including:

[0016] S1: Obtain product comparison Q&A text information from user community Q&A on online platforms, and construct comparison Q&A text containing comparison questions and their answers.

[0017] Specifically, the steps include: acquiring consumer question text data and consumer answer text data from the community Q&A section of the mobile phone product on the online platform, and filtering out comparison questions and their answer texts from the consumer question text data and consumer answer text data based on a large language model, and constructing comparison question and answer texts.

[0018] In this embodiment, examples of consumer question text data and consumer answer text data are shown in Table 1;

[0019] Table 1. Examples of consumer question text data and consumer response text data for 14 mobile phone models of brand A.

[0020] ;

[0021] Furthermore, a comparison question recognition agent is designed based on a large language model and the first prompt word. The comparison questions are then filtered from the text data using the large language model based on the first prompt word. The specific concept is as follows:

[0022] Taking mobile phone brand A as an example, the "questions" in the consumer question text data are questions asked by users about mobile phone brand A.

[0023] First, the problem recognition agent determines whether the "problem" mentions a specific model of a mobile phone product other than brand A. If it does, the output follows the original content of the "problem"; otherwise, the output is empty.

[0024] Secondly, if the consumer question text does not explicitly mention a brand, but the context and wording indicate a comparison question related to phone brand A and involving specific models of other phones, then the original content of the question should be output. For example, "Which is better, 13 or 14?" should be understood as "Which is better, phone brand A's 13 model or 14 model?", and the original question should be output. Similarly, "Which is better, the standard version or the Pro version?" should be understood as "Which is better, phone brand A's standard version model or the Pro version model?", and the original question should be output.

[0025] Furthermore, if the question mentions other brands but not specific models, such as "Which should I buy, mobile phone brand A or mobile phone brand C?", although mobile phone brand C is mentioned, the specific mobile phone model of mobile phone brand C is not mentioned, then the output will be empty.

[0026] The constraints for identifying agents in a comparison problem are as follows:

[0027] 1. Only handle questions related to comparisons of mobile phone products, and refuse to answer questions unrelated to comparisons of mobile phone products.

[0028] 2. Make judgments based solely on the input "problem" and do not change the content of the "problem".

[0029] The workflow of the comparison question recognition agent is as follows: 1) Read data: Obtain all consumer question text data that need to be recognized from the table; 2) Transmit data: Input all consumer question text data that need to be recognized into the large language model of the first prompt word in sequence; 3) Data processing: Filter the input consumer question text data based on the large language model of the first prompt word, and filter out ordinary questions (all other questions except comparison questions); 4) Data storage: Construct comparison question-and-answer text based on the retained comparison questions and their answer texts, and store it in cloud documents.

[0030] S2: Classify all comparison question and answer texts according to the attributes of the mobile phone product, obtain the first comprehensive intuition fuzzy number of the comparison questions corresponding to the attribute based on the first sentiment score agent; calculate the mean of the first comprehensive intuition fuzzy number of all comparison questions corresponding to the attribute to obtain the first fuzzy number of the mobile phone product under the corresponding attribute.

[0031] In this embodiment, attributes include appearance, performance, imaging, and a comprehensive comparison. This is suitable for categorizing questions where users haven't specified any particular phone attributes and are only asking general questions about advantages and disadvantages. (Mobile phone product) In the The set of comparison problems under each attribute is denoted as . ,in, For mobile phone products In the The first attribute under the A comparative question; For mobile phone products In the The number of comparison questions under each attribute; The collection of answer texts is ,in, for The One answer text; It is a mobile phone product. In the The first attribute under the The number of answer texts for each comparison question.

[0032] Based on a large language model and a second cue word, an intelligent agent for attribute recognition in comparison problems is designed. The specific design concept is as follows:

[0033] First, carefully analyze the mobile phone-related questions in the input content of the large language model. Each time, a question will be obtained, regardless of how many sentences the question contains, it will be treated as a question.

[0034] Secondly, the problems can be accurately categorized into the following four types:

[0035] Appearance includes screen display quality, body design, materials, color, size, weight, and physical characteristics such as curved / flat screen. Questions include: Is it drop-resistant? How is the screen clarity? Is it easy to apply a screen protector to a curved screen?

[0036] Performance encompasses processor, memory, storage, heat dissipation, chip, and communication-related hardware (such as signal strength, 5G support, Wi-Fi / Bluetooth, dual SIM functionality, and speaker). Questions include: Is the phone laggy? Is the storage capacity sufficient? Is the signal stronger than the previous generation? It also includes pure software aspects such as operating system functionality, software compatibility, system ads, ecosystem integration, and cross-device interaction. Questions include: Does HarmonyOS have ads? Can Google services be installed?

[0037] Imaging, encompassing camera configuration, photo quality, video recording, macro / night mode capabilities, algorithm optimization, and other photographic capabilities, raises questions such as: Is its photo quality better than an iPhone's? Is its nighttime autofocus faster?

[0038] The comprehensive comparison includes model comparison, brand recommendations, and purchase advice (decision-making issues such as choosing between new and old models and the timing of purchase). Questions include: Which is better, mobile phone brand C or mobile phone brand D? Should I buy now or wait for the new product? If a situation can be categorized into both the "Other Categories" and "Comprehensive Comparison" categories, it will not be categorized into the "Comprehensive Comparison" category, but rather into one of the other three categories.

[0039] The comparison problem attribute recognition agent will classify the comparison problems and output them in the following format: [{"fields":{"attributes of comparison problems":"problem content"}}]. The output content must not be modified in any way.

[0040] The constraints for the attribute recognition agent in the comparison problem are as follows:

[0041] 1. Only handle questions related to mobile phones; refuse to answer questions unrelated to mobile phones.

[0042] 2. The classification results must accurately correspond to the above four categories, and misclassification is not allowed;

[0043] 3. It can only be divided into one category, and the original sentence should not be split.

[0044] Output example: The question "Is taking photos better than mobile phone brand D?" is categorized as "images" and the output is [{"fields":{"images":"Is taking photos better than mobile phone brand D?"}}].

[0045] Examples of comparison questions are shown in Table 2.

[0046] Table 2 Examples of Comparison Problems

[0047] ;

[0048] Specifically, the process of obtaining the first fuzzy number of a mobile phone product under the corresponding attribute includes:

[0049] Based on the first sentiment score agent, obtain the first positive sentiment score and the first negative sentiment score of any answer text corresponding to any attribute;

[0050] Aggregate the first positive sentiment scores of all responses to the same comparison question to obtain the attribute positive sentiment score for the corresponding comparison question; the aggregation formula is:

[0051] ;

[0052] in, Indicates the first A first-level sentiment scoring AI agent for mobile phone products In the The first attribute under the The attributes of the comparison question have positive sentiment scores; Indicates mobile phone products In the The first attribute under the The number of answer texts for each comparison question; Indicates the first A first-level sentiment scoring AI agent for mobile phone products In the The first attribute under the The first comparison question The first positive sentiment score for each response text.

[0053] The first emotional scoring agent for mobile phone product 1 (mobile phone brand A) , , Table 3 shows examples of the positive and negative sentiment scores of the attributes for the third comparison question under different performance conditions.

[0054] Table 3. Data examples of attribute positive sentiment scores and attribute negative sentiment scores.

[0055] ;

[0056] Aggregate the first negative sentiment scores of all responses to the same comparison question to obtain the attribute negative sentiment score for the corresponding comparison question; the aggregation formula is:

[0057] ;

[0058] in, Indicates the first A first-level sentiment scoring AI agent for mobile phone products In the The first attribute under the The negative sentiment score of the attribute of the comparison question; Indicates the first A first-level sentiment scoring AI agent for mobile phone products In the The first attribute under the The first comparison question The first negative sentiment score for each response text.

[0059] The positive sentiment score of the attribute for the same comparison question is used as the first membership degree (indicating the degree to which the target product is superior to the competitors), and the negative sentiment score of the attribute for the same comparison question is used as the first non-membership degree (indicating the degree to which the target product is inferior to the competitors).

[0060] Combining the first membership degree and the first non-membership degree of the same contrast problem, we take the first attribute intuitive fuzzy number of the first sentiment scoring agent for the same contrast problem. The first attribute intuitive fuzzy number is denoted as... ;

[0061] Based on the first aggregation weight, the first attribute intuition fuzzy number of multiple first sentiment scoring agents for the same comparison question is aggregated to obtain the first comprehensive intuition fuzzy number of the comparison question corresponding to the attribute. The first comprehensive intuition fuzzy number is denoted as . The aggregation formula is:

[0062] ;

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] in, Indicates mobile phone products In the The first attribute under the The first comprehensive intuitive fuzzy number of a comparison problem; Represents the first comprehensive intuition fuzzy number The first membership degree; Represents the first comprehensive intuition fuzzy number The first non-membership degree; L represents the number of agents that give the first sentiment score; This indicates the first aggregation weight, which is relevant to the comparison problem; Indicates the first A first-level sentiment scoring AI agent for mobile phone products In the The first attribute under the The first attribute of a comparison problem is intuitive fuzzy numbers; Indicating the first attribute intuitive fuzzy number The mean; Indicates positive sentiment score The mean; Indicates negative sentiment score. The mean.

[0068] The mean of the first comprehensive intuitive fuzzy numbers for all comparison questions corresponding to the attribute is used to obtain the first fuzzy number of the mobile phone product under the corresponding attribute. This first fuzzy number is denoted as... The calculation formula is:

[0069] ;

[0070] ;

[0071] in, Indicates mobile phone products In the The first fuzzy number under each attribute; Represents the first fuzzy number The first membership degree; Indicates mobile phone products In the The number of comparison questions under each attribute; Represents the first fuzzy number The first non-membership degree.

[0072] Furthermore, a first sentiment scoring agent is designed based on a large language model and a third cue word. The specific design concept is as follows:

[0073] First, a detailed analysis of each response text is performed using a large language model to identify the specific evaluation content of the target product and competitors. Second, the first evaluation scores of the target product and competitors are specified to be in the range of [0,1], with the closer the value is to 1, the better the evaluation, and the sum of the first evaluation scores of the target product and competitors is less than or equal to 1.

[0074] Example:

[0075] Question: Which mobile phone brand would you recommend, brand A or brand B? The target product is brand A.

[0076] Answer 1: 14. Therefore, mobile phone brand A scores 0.7, and mobile phone brand B scores 0.2.

[0077] Answer 2: For gaming, choose phone brand B; for light gaming and photography, choose phone brand A; phone brand B is also a good option. Therefore, phone brand A scores 0.5, and phone brand B scores 0.5.

[0078] The first sentiment rating agent is restricted as follows:

[0079] 1. During the scoring process, the sum of the first evaluation scores of the target product and the competitors must be less than or equal to 1.

[0080] 2. The first evaluation score must match the actual evaluation in the answer text. Positive evaluations should receive higher scores, and negative evaluations should receive lower scores.

[0081] 3. Only evaluate and output answers to the comparison questions, and refuse to answer questions on topics unrelated to the task.

[0082] 4. The output must conform to the given format, which is:

[0083] “mb_score”: “Positive sentiment score of the target product’s attributes”;

[0084] "jp_score": "Positive sentiment score of competitor attributes".

[0085] Furthermore, the process of obtaining the first positive sentiment score and the first negative sentiment score includes:

[0086] For comparison questions that belong to the same attribute, the corresponding answer text

[0087] Analyze the first evaluation scores of the target product and the competitor in any answer text of the same comparison question. The first evaluation score is a number in the range [0,1]. The closer the value is to 1, the better the evaluation. The sum of the first evaluation scores of the target product and the competitor is less than or equal to 1.

[0088] The first evaluation score of the target product in any answer text of the same comparison question is taken as the first positive sentiment score, and the first evaluation score of the corresponding competitor is taken as the first negative sentiment score.

[0089] S3: Obtain the complete video content of the corresponding mobile phone product evaluation and comparison video, and extract the video text information of the complete video content.

[0090] Specifically, the steps include: obtaining the title, URL, basic information of the corresponding blogger, and interaction information of the review and comparison videos for the corresponding mobile phone products; performing preliminary screening of the review and comparison videos based on the interaction information and the basic information of the corresponding blogger; inputting the URL of the preliminarily screened review and comparison videos into the plugin tool of the Coze platform to obtain the corresponding complete video content; and extracting the video text information of the complete video content based on the video speech-to-text plugin in the large model platform and conventional multimodal information fusion technology.

[0091] In this embodiment, the blogger's basic information includes the number of followers, total likes, and verification status, which are represented as follows: , , (When already verified) When not certified Interactive information includes the number of likes, favorites, comments, and shares of the evaluation and comparison videos, expressed as follows: , , , .

[0092] This embodiment provides some data examples from the evaluation and comparison videos related to 14 models of mobile phone brand A, as shown in Table 4;

[0093] Table 4. Examples of partial data from comparison videos of 14 mobile phone models related to mobile phone brand A.

[0094] ;

[0095] Furthermore, the initial screening of the evaluation and comparison videos includes:

[0096] For verified bloggers, their review and comparison videos will be retained directly;

[0097] For unverified bloggers, review and comparison videos from bloggers with fewer than 100,000 followers and fewer than 500,000 total likes are excluded.

[0098] Furthermore, comparison videos with fewer than 1,000 likes in the interactive information were excluded.

[0099] S4: Split the video text information into multiple text parts according to the attributes, and obtain the second comprehensive intuition fuzzy number of the text part corresponding to the attribute based on the second sentiment scoring agent; aggregate the second comprehensive intuition fuzzy numbers of all text parts corresponding to the attribute to obtain the second fuzzy number of the mobile phone product under the corresponding attribute.

[0100] Specifically, the steps include:

[0101] For any given video text information, it is split into multiple text parts according to appearance, performance, and image quality. In this embodiment, the splitting is implemented based on an attribute classification agent, which is designed based on a large language model and a fourth cue word. The specific concept is as follows:

[0102] Accept a text message containing a video comparing the target product and its competitors.

[0103] The video text information is divided into three parts:

[0104] - Appearance: Covers screen display effects, body design, materials, colors, size, weight, curved / flat screen, and other physical characteristics. The separated appearance content is output to... .

[0105] - Performance: This includes the processor, memory, storage, heat dissipation, chips, and communication-related hardware (such as signal strength, 5G support, Wi-Fi / Bluetooth, dual SIM functionality, and speakers). The broken-down performance data is output to... .

[0106] - Imaging: Includes camera configuration, photo effects, video recording, macro / night scene functions, algorithm optimization, and other photographic capabilities. The split image content is output to... .

[0107] The evaluation criteria are:

[0108] - Only when the text contains clear evaluation conclusions, specific comparison results, or clear evaluations of advantages and disadvantages is it considered a valid evaluation result;

[0109] - Merely mentioning comparisons or describing test scenarios without providing actual results is not considered a valid evaluation result. Valid evaluation results must include explicit comparison statements (such as "A is better than B", "A's effect is clearer", etc.) or specific performance / effect descriptions;

[0110] - If a category (appearance / performance / image) does not contain valid evaluation results, the output for that category will be empty.

[0111] A constrained attribute-classifying agent has the following constraints:

[0112] 1. Only process content related to the text breakdown of the target product and competitor comparison evaluations; refuse to answer irrelevant questions.

[0113] 2. The output content must be organized according to the given format and must not deviate from the framework requirements.

[0114] 3. The output content must be entirely derived from the video text information and no modifications are allowed.

[0115] The output results must strictly follow the following format:

[0116] "waiguan": Appearance evaluation text;

[0117] “xingneng”: Performance evaluation text;

[0118] “yingxiang”: The evaluation text of the image.

[0119] The second sentiment scoring agent obtains the second positive sentiment score and the second negative sentiment score for the text portion corresponding to any attribute. The second sentiment scoring agent is designed based on a large language model and a fifth cue word, and its specific concept is as follows:

[0120] First, extract the video text information, extracting key information from “waiguan”, “xingneng”, and “yingxiang” respectively. Second, analyze the second evaluation scores of the target product and competitors in the same text section. The second evaluation score is a number in the range [0,1]. The closer the value is to 1, the better the evaluation. When the sum of the second evaluation scores of the target product and competitors is less than or equal to 1, and the text section corresponding to the attribute is empty, the second evaluation scores of the target product and competitors corresponding to the attribute are both 0.

[0121] Rating example: Appearance comparison,

[0122] Phone brand A's model 14 adopts the "Light and Shadow Sculpture" design language, with an AG frosted glass back that resists fingerprints, an aviation-grade aluminum alloy frame, and weighs 193g. The four-curved screen has a natural transition, a high screen-to-body ratio, and a stunning visual effect. Phone brand D's model 15 features a highly recognizable dynamic island design, but is lighter, and its screen transition and screen-to-body ratio are not as good as phone brand A's model 14.

[0123] The second evaluation scores are respectively = 0.55, = 0.4.

[0124] The second sentiment scoring agent is restricted as follows:

[0125] 1. The analysis and scoring are based solely on the input video-to-text content and do not involve other unrelated topics.

[0126] 2. When there is uncertainty or both products have common shortcomings, the sum of the scores should be less than 1 to reflect this uncertainty.

[0127] 3. The output should be concise and clear, accurately presenting the scoring results.

[0128] 4. Ensure that when any input (appearance, performance, image) is empty or null, the corresponding score must be 0.

[0129] 5. Strictly ensure that the sum of the scores of the target product and competitors does not exceed 1 in each attribute category (appearance, performance, image).

[0130] The output results must strictly follow the following format:

[0131] “mb_wg”: Target product appearance score

[0132] "jp_wg": Competitor's appearance score

[0133] “mb_xn”: Target product performance score

[0134] "jp_xn": Competitor performance score

[0135] “mb_yx”: Image score of the target product

[0136] “jp_yx”: Competitor image score.

[0137] The second positive sentiment score of the same text part is used as the second membership degree, and the second negative sentiment score of the same text part is used as the second non-membership degree; the second membership degree and the second non-membership degree of the same text part are combined as the second sentiment score agent's second attribute intuition fuzzy number for the same text part;

[0138] Based on the second aggregation weight, the second attribute intuition fuzzy number of multiple second sentiment scoring agents on the same text portion is aggregated to obtain the second comprehensive intuition fuzzy number of the text portion corresponding to the attribute. The second comprehensive intuition fuzzy number is denoted as... The aggregation formula is:

[0139] ;

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] in, Indicates the first A comparison video of mobile phone products In the The second comprehensive intuition fuzzy number under each attribute; Representing the second synthetic intuition fuzzy number The second membership degree; Representing the second synthetic intuition fuzzy number The second non-membership degree; M represents the number of agents for the second sentiment rating; This indicates the second aggregation weight, which applies to the text portion corresponding to the evaluation comparison video. Indicates the first The second emotion scoring agent on the first A comparison video of mobile phone products In the The intuitive fuzzy number of the second attribute under each attribute; Indicating the second attribute intuitive fuzzy number The mean; Indicating the second attribute intuitive fuzzy number The second membership degree; express The mean; Indicating the second attribute intuitive fuzzy number The second non-membership degree; express The mean.

[0145] Second-level sentiment scoring agent , , Table 5 shows examples of the second membership degree and second non-membership degree data of the second attribute intuitive fuzzy number of mobile phone product 1 under the fourth evaluation comparison video.

[0146] Table 5. Examples of second membership and second non-membership data for second attribute intuitive fuzzy numbers.

[0147] ;

[0148] Traverse all video text information to find the text parts that belong to the same attribute, and obtain the second comprehensive intuition fuzzy number of all text parts corresponding to the attribute;

[0149] The second comprehensive intuitive fuzzy number of all text parts corresponding to the aggregated attribute is used to obtain the second fuzzy number of the mobile phone product under the corresponding attribute. The second fuzzy number is denoted as... The aggregation formula is:

[0150] ;

[0151] ;

[0152] in, Indicates mobile phone products In the The second fuzzy number under each attribute; Indicates mobile phone products No. The second membership degree of the second fuzzy number under each attribute; Indicates mobile phone products No. The second non-membership degree of the second fuzzy number under each attribute; H represents the number of comparison videos in the evaluation; This represents the third aggregation weight.

[0153] The formula for calculating the third aggregation weight is:

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] ;

[0163] ;

[0164] ;

[0165] in, Indicates the first A comparison video of mobile phone products The overall score; This represents the preference adjustment parameter, indicating the user's... and The bias in these two dimensions of video information; This represents the average score corresponding to each blogger's information; This represents the average score corresponding to each interactive message; , , These represent the blogger's number of followers, total number of likes, and verification status, respectively. , , They represent , , The corresponding score; , , , These represent the number of likes, favorites, comments, and shares for the evaluation and comparison videos, respectively. , , , They represent , , , The corresponding score.

[0166] Furthermore, the process of obtaining the second positive sentiment score and the second negative sentiment score includes:

[0167] For any text portion belonging to the same attribute,

[0168] Analyze the second evaluation scores of the target product and competitors in the same text section. The second evaluation score is a number in the range [0,1]. The closer the value is to 1, the better the evaluation. When the sum of the second evaluation scores of the target product and competitors is less than or equal to 1 and the text section corresponding to the attribute is empty, the second evaluation scores of the target product and competitors corresponding to the attribute are both 0.

[0169] The second evaluation score of the target product in the same text section is used as the second positive sentiment score, and the second evaluation score of the corresponding competitor is used as the second negative sentiment score.

[0170] S5: Based on the first and second fuzzy numbers under all attributes, calculate the recommendation score of the mobile phone product, and select the mobile phone product with the highest recommendation score as the recommended product.

[0171] Specifically, the steps include:

[0172] Aggregate the first and second fuzzy numbers under all attributes;

[0173] The aggregation formula for the first fuzzy number under all attributes is:

[0174] ;

[0175] ;

[0176] in, Indicates mobile phone products The first membership degree of the first fuzzy number after aggregation; This indicates the adjustment parameter. A larger value indicates that the user is more concerned about a particular outstanding feature of the product (appearance / performance / image), while a smaller value indicates that the user is more concerned about the overall performance of the product (comprehensive comparison). Indicates the first Personalized weights for the first or second fuzzy number of each attribute; Indicates mobile phone products No. The first membership degree of the first fuzzy number under each attribute; Indicates mobile phone products The first membership degree of the first fuzzy number under comprehensive comparison; Indicates mobile phone products The first non-membership degree of the first fuzzy number after aggregation; Indicates mobile phone products No. The first non-membership degree of the first fuzzy number under each attribute; Indicates mobile phone products The first non-membership degree of the first fuzzy number under comprehensive comparison;

[0177] The aggregation formula for the second fuzzy number under all attributes is:

[0178] ;

[0179] ;

[0180] in, Indicates mobile phone products The second membership degree of the corresponding aggregated second fuzzy number; Indicates mobile phone products No. The second membership degree of the second fuzzy number under each attribute; Indicates mobile phone products The second non-membership degree of the corresponding aggregated second fuzzy number; Indicates mobile phone products No. The second non-membership degree of the second fuzzy number under each attribute;

[0181] Based on the attention weights, the first and second fuzzy numbers are aggregated again to obtain the final comprehensive intuitionistic fuzzy number; the aggregation formula is:

[0182] ;

[0183] ;

[0184] in, Indicates mobile phone products The membership degree of the final comprehensive intuitionistic fuzzy number; This indicates the weight users place on community Q&A and comparison review videos; Indicates mobile phone products The final non-membership degree of the comprehensive intuitionistic fuzzy number.

[0185] This embodiment provides specific data examples for each hyperparameter, as shown in Table 6;

[0186] Table 6. Specific data examples for each hyperparameter

[0187] ;

[0188] In Table 6, the parameters , , These represent the personalized weights of the first or second fuzzy number for the appearance, performance, and imaging attributes of the aggregated mobile phone product, respectively.

[0189] The recommendation score for a mobile phone product is obtained by subtracting the corresponding non-membership degree from the membership degree of the final comprehensive intuitive fuzzy number.

[0190] Iterate through all mobile phone products, obtain the recommendation score for each product, and recommend the mobile phone product with the highest recommendation score to the user.

[0191] This embodiment provides recommendation scores and ranking results for three mobile phone products, as shown in Table 7.

[0192] Table 7. Examples of Recommendation Scores and Ranking Results for Three Mobile Phone Products

[0193] ;

[0194] As can be seen from Table 7, mobile phones In other words, the X100 model from brand E is the best choice, and it is recommended to users as a recommended product.

[0195] The mobile product recommendation method based on comparing question-and-answer text and comparing videos provided in this embodiment has the following beneficial effects:

[0196] This method improves the accuracy and comprehensiveness of recommendations by integrating two types of information resources: community Q&A and comparison evaluation videos. It recommends mobile phone products that better meet users' needs and are more valuable for reference, helping users to accurately select the right product from a wide range of mobile phone models.

[0197] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0198] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for recommending mobile phone products based on comparative question-and-answer text and comparative videos, characterized in that, include: S1: Obtain product comparison Q&A text information from user community Q&A on online platforms, and construct comparison Q&A text containing comparison questions and their answers; S2: Classify all comparison question and answer texts according to the attributes of the mobile phone product, obtain the first comprehensive intuition fuzzy number of the comparison questions corresponding to the attributes based on the first sentiment score agent; calculate the mean of the first comprehensive intuition fuzzy number of all comparison questions corresponding to the attributes to obtain the first fuzzy number of the mobile phone product under the corresponding attributes; S3: Obtain the complete video content of the evaluation and comparison video of the corresponding mobile phone product, and extract the video text information of the complete video content; S4: Split the video text information into multiple text parts according to the attributes, and obtain the second comprehensive intuition fuzzy number of the text part corresponding to the attribute based on the second sentiment scoring agent; aggregate the second comprehensive intuition fuzzy numbers of all text parts corresponding to the attribute to obtain the second fuzzy number of the mobile phone product under the corresponding attribute. S5: Based on the first and second fuzzy numbers under all attributes, calculate the recommendation score of the mobile phone product, and select the mobile phone product with the highest recommendation score as the recommended product.

2. The mobile product recommendation method based on comparative question-and-answer text and comparative video as described in claim 1, characterized in that, S1 includes: Obtain text data of consumer questions and answers from community Q&A sessions for mobile phone products on online platforms, and use a large language model to filter out comparison questions and their answers from the text data of consumer questions and answers to construct comparison Q&A text.

3. The mobile product recommendation method based on comparative question-and-answer text and comparative video as described in claim 1, characterized in that, The attributes include appearance, performance, imaging, and overall comparison.

4. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 3, characterized in that, In S2, the process of obtaining the first fuzzy number of the mobile phone product under the corresponding attribute includes: Based on the first sentiment score agent, obtain the first positive sentiment score and the first negative sentiment score of any answer text corresponding to any attribute; Aggregate the first positive sentiment scores of all answer texts for the same comparison question to obtain the attribute positive sentiment scores for the corresponding comparison question; Aggregate the first negative sentiment scores of all answer texts for the same comparison question to obtain the attribute negative sentiment score for the corresponding comparison question; The positive sentiment score of the attribute for the same comparison question is used as the first membership degree, and the negative sentiment score of the attribute for the same comparison question is used as the first non-membership degree. The first membership degree and the first non-membership degree of the same comparison question are combined as the first attribute intuitive fuzzy number of the first sentiment scoring agent for the same comparison question; Based on the first aggregation weight, the first comprehensive intuition fuzzy number of the first attribute of the comparison question is obtained by aggregating the first attribute intuition fuzzy number of multiple first sentiment scoring agents for the same comparison question. The mean of the first comprehensive intuitionistic fuzzy numbers for all comparison questions corresponding to the attribute is used to obtain the first fuzzy number of the mobile phone product under the corresponding attribute.

5. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 4, characterized in that, The process of obtaining the first positive sentiment score and the first negative sentiment score includes: For comparison questions that belong to the same attribute, the corresponding answer text Analyze the first evaluation scores of the target product and the competitor in any answer text of the same comparison question. The first evaluation score is a number in the range [0,1]. The closer the value is to 1, the better the evaluation. The sum of the first evaluation scores of the target product and the competitor is less than or equal to 1. The first evaluation score of the target product in any answer text of the same comparison question is taken as the first positive sentiment score, and the first evaluation score of the corresponding competitor is taken as the first negative sentiment score.

6. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 3, characterized in that, S3 include: The system retrieves the titles, URLs, basic information of the corresponding bloggers, and interaction information of the review and comparison videos for the corresponding mobile phone products. Based on the interaction information and the basic information of the corresponding bloggers, the review and comparison videos are initially screened. The URLs of the initially screened review and comparison videos are then input into the plugin tool of the Coze platform to obtain the corresponding complete video content. The video text information of the complete video content is then extracted based on the video speech-to-text plugin in the large model platform and conventional multimodal information fusion technology.

7. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 6, characterized in that, The initial screening of the comparison videos includes: For verified bloggers, their review and comparison videos will be retained directly; For unverified bloggers, review and comparison videos from bloggers with fewer than 100,000 followers and fewer than 500,000 total likes are excluded. Furthermore, comparison videos with fewer than 1,000 likes in the interactive information were excluded.

8. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 3, characterized in that, S4 include: For any given video text information, it is broken down into multiple text parts according to appearance, performance, and image quality; The second sentiment score agent obtains the second positive sentiment score and the second negative sentiment score of the text portion corresponding to any attribute. The second positive sentiment score of the same text part is used as the second membership degree, and the second negative sentiment score of the same text part is used as the second non-membership degree; the second membership degree and the second non-membership degree of the same text part are combined as the second sentiment score agent's second attribute intuition fuzzy number for the same text part; Based on the second aggregation weight, multiple second sentiment scoring agents aggregate the second attribute intuition fuzzy numbers of the same text part to obtain the second comprehensive intuition fuzzy number of the text part corresponding to the attribute. Traverse all video text information to find the text parts that belong to the same attribute, and obtain the second comprehensive intuition fuzzy number of all text parts corresponding to the attribute; The second comprehensive intuitive fuzzy number of all text parts corresponding to the aggregated attribute is used to obtain the second fuzzy number of the mobile phone product under the corresponding attribute.

9. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 8, characterized in that, The process of obtaining the second positive sentiment score and the second negative sentiment score includes: For any text portion belonging to the same attribute, Analyze the second evaluation scores of the target product and competitors in the same text section. The second evaluation score is a number in the range [0,1]. The closer the value is to 1, the better the evaluation. When the sum of the second evaluation scores of the target product and competitors is less than or equal to 1 and the text section corresponding to the attribute is empty, the second evaluation scores of the target product and competitors corresponding to the attribute are both 0. The second evaluation score of the target product in the same text section is used as the second positive sentiment score, and the second evaluation score of the corresponding competitor is used as the second negative sentiment score.

10. The mobile product recommendation method based on comparative question-and-answer text and comparative video according to claim 3, characterized in that, S5 include: Aggregate the first and second fuzzy numbers under all attributes; The aggregation formula for the first fuzzy number under all attributes is: ； ； in, Indicates mobile phone products The first membership degree of the first fuzzy number after aggregation; Indicates the adjustment parameter; Indicates the first Personalized weights of the first or second fuzzy number of each attribute; Indicates mobile phone products No. The first membership degree of the first fuzzy number under each attribute; Indicates mobile phone products The first membership degree of the first fuzzy number under comprehensive comparison; Indicates mobile phone products The first non-membership degree of the first fuzzy number after aggregation; Indicates mobile phone products No. The first non-membership degree of the first fuzzy number under each attribute; Indicates mobile phone products The first non-membership degree of the first fuzzy number under comprehensive comparison; The aggregation formula for the second fuzzy number under all attributes is: ； ； in, Indicates mobile phone products The second membership degree of the corresponding aggregated second fuzzy number; Indicates mobile phone products No. The second membership degree of the second fuzzy number under each attribute; Indicates mobile phone products The second non-membership degree of the corresponding aggregated second fuzzy number; Indicates mobile phone products No. The second non-membership degree of the second fuzzy number under each attribute; Based on the attention weights, the first and second fuzzy numbers are aggregated again to obtain the final comprehensive intuitionistic fuzzy number; the aggregation formula is: ； ； in, Indicates mobile phone products The membership degree of the final comprehensive intuitionistic fuzzy number; This indicates the weight users place on community Q&A and comparison review videos; Indicates mobile phone products The non-membership degree of the final comprehensive intuitionistic fuzzy number; The recommendation score for a mobile phone product is obtained by subtracting the corresponding non-membership degree from the membership degree of the final comprehensive intuitive fuzzy number. Iterate through all mobile phone products, obtain the recommendation score for each product, and recommend the mobile phone product with the highest recommendation score to the user.

Images