An Interest Data Recognition Method, System and Medium for an AI Intelligent Mobile Terminal

Abstract

The present invention discloses a method, system and medium for identifying interest data of an AI intelligent mobile terminal, specifically related to the field of artificial intelligence, including S1: information acquisition, S2: user usage evaluation management, S3: interest tag acquisition, S4: user preference tag acquisition, S5: advertisement similarity recognition, and S6: advertisement recommendation display. The present invention acquires user information through a mobile terminal and manages it based on user usage evaluations, not limited to basic data analysis techniques that only perform simple statistics. After acquiring interest tags and user preference tags, it evaluates the similarity of preferences to achieve a high-precision grasp of the delivery volume of personalized advertisements, avoiding waste of resources and increase in costs caused by unnecessary advertisement delivery. At the same time, when the user's demand for matching advertisements decreases, the user can switch to the optimal options after screening and sorting through an interactive hotkey, saving the user's time and energy and improving the matching efficiency between advertisements and demands.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to a method, system and medium for identifying interest data of an AI smart mobile terminal. Background Art

[0002] Against the backdrop of the deep integration of the digital economy and the real economy, mobile terminals have become the core carrier of user behavior data. The average daily usage time of users is as high as 9-12 hours. Mobile phones are increasingly becoming the largest traffic entrance and source of information data collection. Various user behaviors on mobile terminals form massive and multi-dimensional interest-related data, providing a data foundation for accurately identifying user interests and realizing personalized services.

[0003] At present, in the field of interest data identification on mobile terminals, the main existing technologies used include classification methods based on simple rules and some basic data analysis technologies. The classification methods based on simple rules are usually based on pre-set fixed rules. For example, users’ interests are preliminarily judged based on the specific websites they browse or the specific types of apps they use. The basic data analysis technology performs simple statistical analysis on the collected data, such as calculating the frequency of users’ browsing of various types of content, in order to infer users’ interest tendencies. In addition, some companies also obtain interest data through methods such as users manually filling out questionnaires and selecting interest tags.

[0004] However, it still has some shortcomings in actual use. First, the classification method based on simple rules relies too much on preset rules and cannot fully and deeply understand the real interests of users. It only judges interests based on browsing specific websites, ignoring the specific behaviors and diverse interest expressions of users on the website. The basic data analysis technology only performs simple statistics and it is difficult to mine the complex associations and potential interests behind the data, resulting in inaccurate judgment of user interests and inability to achieve accurate content recommendation and marketing.

[0005] Second, in actual applications, users' interests are complex and changeable, and are affected by many factors. In this case, it is difficult to comprehensively consider these factors and accurately identify user interests in different scenarios. Users' interests may be different in different time periods and geographical locations, but the existing methods cannot effectively capture such changes, resulting in user interest identification lagging behind actual needs. Summary of the invention

[0006] In view of this, embodiments of the present invention provide a method, system, and medium for identifying interest data of an AI intelligent mobile terminal. By obtaining user information through the mobile terminal, further managing the user's usage evaluation, and making refined and flexible adjustments, it maximally ensures the combination of the user's usage experience and recommendation adjustment. At the same time, it displays advertisement recommendations, effectively solving the problems raised in the background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] S1: Information acquisition: Set an advertisement entry after the mobile terminal is turned on the screen, obtain the user's basic information through the advertisement entry, where the user's basic information is the user's basic personal profile and information about their consumption situation, and allow the user to upload a usage evaluation on the advertisement details page;

[0009] S2: User usage evaluation management: Identify valid evaluations based on the user's usage evaluation, and at the same time perform core vocabulary recognition on the valid evaluations and conduct score classification, where the score classification includes positive scores and negative scores;

[0010] S3: Interest tag acquisition: Extract the proportion of positive scores based on user usage evaluation management, and classify advertisement categories, where the advertisement categories include positive advertisements and negative advertisements. Analyze the deviation degree of advertisement placement according to the advertisement categories, and thus generate user interest tags;

[0011] S4: User preference tag acquisition: Obtain the user's behavior data through the mobile terminal, analyze the user preference coefficient corresponding to each advertisement when the user uses it according to the user's behavior data, and judge the degree of user preference based on the user preference coefficient, and thus generate user preference tags;

[0012] S5: Advertisement similarity recognition: Perform similarity matching based on the user interest tags and user preference tags, evaluate the advertisement similarity accordingly, and sort each advertisement in a preset order according to the evaluation result;

[0013] S6: Advertisement recommendation display: Display personalized advertisements in real time based on the sorting result, trigger advertisement switching through hotkey interaction or gesture operation, and update the tags and sorting according to the user's real-time behavior feedback.

[0014] The technical effects and advantages of the present invention:

[0015] 1. The present invention obtains user information through a mobile terminal and manages it based on user usage evaluations, thereby obtaining interest tags and user preference tags. It is not limited to basic data analysis techniques that only perform simple statistics. On the one hand, it can deeply match interest data with advertisements. On the other hand, it can avoid the one-sidedness of ignoring negative signals due to positive evaluations of advertisements by users, and improve the accuracy of user satisfaction recognition;

[0016] 2. The present invention determines that the advertisement ranked first is the one that matches the user's preferences. The matching advertisement can generate eye-catching titles, copywriting suggestions, and video scripts through AI, improving content quality and production efficiency. It reaches users efficiently by unlocking the screen and reducing interference. The matching advertisement ranked first is preferentially displayed on the startup screen to ensure that the user first sees the advertisement they are most interested in, thereby arousing the user's interest and increasing the user's click-through rate. At the same time, when the user's demand for the matching advertisement decreases, they can switch to the optimal option after screening and sorting through an interactive hotkey, saving the user's time and energy and improving the matching efficiency between advertisements and demands. On the other hand, since forced full-screen advertisements are likely to arouse user disgust, providing an interactive hotkey to switch advertisements gives users control, enabling them to more conveniently view other matching advertisements they are interested in, thereby increasing the depth of user interaction and increasing the user's stay time and engagement;

[0017] 3. After obtaining interest tags and user preference tags, the present invention evaluates the similarity of preferences, thereby performing advertisement ranking, achieving high-precision control of the delivery volume of personalized advertisements, and avoiding waste of resources and increase in costs caused by unnecessary advertisement delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is the overall structural flowchart of the present invention.

[0019] Figure 2 is the overall structural schematic diagram of the present invention.

[0020] Figure 3 is the flowchart for obtaining interest tags of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] As shown in the appended Figure 1 A method for identifying interest data of an AI intelligent mobile terminal shown, the specific implementation method includes the following steps:

[0023] S1: Information acquisition: Set an advertisement entry after the mobile terminal is powered on. Obtain the user's basic information through the advertisement entry, where the user's basic information is the user's basic personal profile and information about their consumption situation, and allow the user to upload a usage evaluation on the advertisement details page.

[0024] In this embodiment, it should be specifically noted that the user's basic information generally refers to the user's basic personal profile and information about their consumption situation in interest data recognition, including the user's gender, age, historical consumption times and items, and the category of browsing content. The user's basic information will represent the user's specific situation, and the user will be classified according to these situations. The user classification can specifically be the classification of the user's age group. For example, the user classification situations are 18 - 24 years old, 24 - 35 years old, and 35 - 44 years old.

[0025] It should be further noted that the splash screen advertisement reaches the user through the advertisement entry, supports one - click jump to the advertisement details page. By setting a function integration unit on the advertisement details page of the mobile terminal, such as "collect" or "share" an item, the user is thus guided to upload a usage evaluation, where the usage evaluation includes the evaluation content and evaluation star rating of historical users.

[0026] It should be explained that before obtaining the user's basic information, it is necessary to clearly inform the data collection scope and purpose in the advertisement entry pop - up window. For example, a authorization pop - up window is popped up to request consent for the advertisement entry and data collection, so as to reduce the risk of privacy leakage and improve privacy protection.

[0027] S2: User usage evaluation management: Identify valid evaluations according to the user's usage evaluation, and at the same time perform core vocabulary recognition on the valid evaluations and conduct score division, where the score division includes positive scores and negative scores.

[0028] In this embodiment, it should be specifically noted that the specific method for identifying valid evaluations is as follows:

[0029] Segment, clean, and divide the evaluation content of each usage evaluation, and perform part - of - speech tagging on the segments.

[0030] It should be explained that segment cleaning and division can be carried out through a Chinese word segmentation tool to remove stop words, punctuation marks, and irrelevant characters, such as "le", "he", "de". Part - of - speech tagging can be performed through a part - of - speech tagging tool, and the part - of - speech includes but is not limited to nouns and adjectives.

[0031] Identify core vocabulary based on the part - of - speech of each segment through a predefined sentiment word library. If no core vocabulary is identified in the usage evaluation, then eliminate this usage evaluation.

[0032] In the preferred implementation of the above solution, the core vocabulary is identified as follows: adjectives are set as sentiment descriptors, and nouns are set as topic words. Thus, sentiment descriptors and topic words are extracted from the part-of-speech of each segmented word, and adjacent sentiment descriptors and topic words are combined into core vocabulary.

[0033] It should be explained that adjectives are set as sentiment descriptors because adjectives mainly undertake the functions of describing and modifying nouns in language and naturally carry emotional or evaluative attributes. By extracting adjectives, the emotional tendency of users towards a certain object can be directly captured; nouns are set as topic words because nouns are specific objects in language, and in the use of evaluation, nouns can provide detailed information about the evaluation object. As topic words, nouns can clarify the specific content of the emotional orientation; by combining sentiment descriptors and topic words, the focus of user feedback can be accurately positioned, making the evaluation more expressive and persuasive.

[0034] Exemplarily, if the usage evaluation is "This mobile phone, although the price is a bit high, but the performance is very powerful.", the segmented words divided by the Chinese word segmentation tool include "this", "mobile phone", "although", "price", "a", "bit", "high", "but", "performance", "very", "powerful". The segmented words after cleaning include "mobile phone", "price", "high", "performance", "powerful". Using the part-of-speech tagging tool for part-of-speech tagging, among which the nouns include "mobile phone", "price", "performance", and the adjectives include "high", "powerful", then the core vocabulary combined by adjacent sentiment descriptors and topic words is "the price of the mobile phone is high" and "the performance is powerful".

[0035] It should be further explained that the extraction of the user's effective score is as follows: the corresponding effective evaluation star rating is obtained according to the effective evaluation. The effective evaluation star rating is the star rating when the user makes a usage evaluation, including five stars. Among them, the effective score corresponding to one star of the evaluation star rating is 1. Thus, the user's effective score is extracted. If the user has multiple effective evaluations for this advertisement, the average value of the effective evaluation star ratings corresponding to the multiple effective evaluations is calculated as the user's effective score.

[0036] The user's effective score is compared with the preset positive score. The preset positive score is set through the positive feedback of each platform and is set in advance. For example, on an e-commerce platform, usually 3 stars are regarded as the boundary between positive feedback and negative feedback. If the user's effective score is greater than or equal to the preset positive score, it indicates that the user has a good evaluation of this advertisement, then this effective score is used as the positive score. If the user's effective score is less than the preset positive score, it indicates that the user has a poor evaluation of this advertisement, then this effective score is used as the negative score.

[0037] It should be noted that if the user does not make any usage evaluation of the advertisement, the average effective score of the corresponding age group according to the user classification is used as their effective score, so as to provide more accurate recommendations for new users.

[0038] S3: Obtaining interest tags: Based on the user usage evaluation management, extract the proportion of positive scores and divide the advertisement categories, where the advertisement categories include positive advertisements and negative advertisements. Analyze the advertisement placement deviation according to the advertisement categories, and thus generate user interest tags.

[0039] In this embodiment, it should be specifically noted that the advertisement extraction process is as follows: Count the valid evaluations uploaded by the user, extract the proportion of positive scores therefrom, and set the corresponding standard proportion. Exemplarily, the positive standard proportion is 70%. Extract the positive scores that reach the standard proportion, and set the advertisements corresponding to the positive scores that reach the standard proportion as positive advertisements. Set the advertisements corresponding to the positive scores that do not reach the standard proportion and negative scores as negative advertisements. At the same time, mark the user scores corresponding to the negative advertisements as negative scores;

[0040] Respectively extract positive scores and negative scores from the valid evaluations of positive advertisements and negative advertisements, and calculate the positive scores and negative scores corresponding to positive advertisements and negative advertisements to obtain the average positive score and average negative score of each advertisement. At the same time, obtain the median effective score by comparing the valid scores of the advertisements;

[0041] Subtract the average positive score and average negative score of each advertisement from the median effective score, take the absolute value, and then divide by the median effective score to obtain the positive score deviation degree corresponding to the positive advertisement and the negative score deviation degree corresponding to the negative advertisement;

[0042] Based on the comparison of the positive score deviation degree corresponding to the positive advertisement and the negative score deviation degree corresponding to the negative advertisement, obtain the advertisement placement deviation degree, which is specifically expressed as:

[0043] ,

[0044] where De represents the advertisement placement deviation degree, Po represents the positive score deviation degree, Ne represents the negative score deviation degree. The lower the positive score deviation degree, the closer the positive score is to the median of all advertisement valid scores, that is, the positive evaluations are more concentrated in the medium level. The higher the negative score deviation degree, the farther the negative score is from the median, then the more reasonable the advertisement placement, and the smaller the advertisement placement deviation degree;

[0045] The advertising delivery deviation of each advertisement is compared with the specified deviation. For example, the specified deviation is 0.2. If the advertising delivery deviation of an advertisement reaches the specified deviation, it indicates that the average effective score of the advertisement delivery is significantly different from the median effective score, reflecting that the matching degree of the advertisement delivery is poor. In this case, the advertisement needs to be eliminated. If the advertising delivery deviation does not reach the specified deviation, it indicates that the average effective score of the advertisement delivery is slightly different from the median effective score, reflecting that the matching degree of the advertisement delivery is good. If the identified advertisement is a positive advertisement, the positive advertisement is set as a reference advertisement. If the identified advertisement is a negative advertisement, the negative advertisement is set as a related advertisement.

[0046] It needs to be explained that by obtaining the corresponding rated advertisements through the proportion of positive and negative ratings, the rated advertisements are obtained based on valid ratings, which can improve the reliability of recommendation results and user satisfaction; using the median to calculate the median effective rating can reduce the interference of extreme values on the deviation calculation. Generally, most advertisements are concentrated on the medium rating, and the median effective rating calculation can better reflect the user's satisfaction with the advertisement. The average values of the ratings of positive and negative advertisements are calculated respectively to form a symmetrical evaluation of positive and negative samples, avoiding the one-sidedness of traditional methods that only focus on positive feedback and ignore negative signals, thereby improving the accuracy of user satisfaction identification.

[0047] It should be further explained that user interest tags are generated based on reference ads and related ads, specifically by identifying keywords in the reference ads and related ads, and converting the keywords into explainable interest tags. The keyword tag conversion can be specifically refined through a preset tag library. For example, the results of keyword identification based on the reference ads and related ads are "sports shoes" and "fitness mats", and the result of interpretable interest tag conversion of the keywords is "sports enthusiasts", where the preset tag library is integrated according to the product classification tags in the e-commerce platform and social media.

[0048] S4: Obtaining user preference tags: Obtaining user behavior data through mobile terminals, and analyzing the user preference coefficients corresponding to each advertisement when the user uses it based on the user behavior data, judging the user preference degree based on the user preference coefficient, and thus generating user preference tags.

[0049] In this embodiment, it should be specifically explained that the user behavior data is obtained through the mobile terminal, and the user preference coefficient corresponding to each advertisement when the user uses it is analyzed according to the user behavior data. The specific analysis method is as follows: the click rate, page dwell time and repurchase rate of each advertisement of the user are extracted from the mobile terminal and integrated into the user behavior data;

[0050] Extract the longest and shortest page residence times based on the user's page residence time, and compare the difference between the page residence time of this advertisement and the shortest page residence time with the difference between the longest page residence time and the shortest page residence time to obtain the residence time standard value;

[0051] Perform a weighted average based on the click-through rate, residence time standard value, and repurchase rate corresponding to each advertisement, and thus statistically calculate the user preference coefficient. Among them, the higher the user's click-through rate, the higher the residence time standard value, and the higher the repurchase rate, the higher the user preference coefficient;

[0052] Compare the user preference coefficient with the defined preference coefficient. The defined preference coefficient is set according to the user preference feedback of each platform. If the user preference coefficient is greater than or equal to the defined preference coefficient, it means that the user has a higher preference for this advertisement, then set this advertisement as a potential advertisement. If the user preference coefficient is less than the defined preference coefficient, it means that the user has a lower preference for this advertisement, then exclude this advertisement.

[0053] It should be explained that the user's click-through rate on the advertisement reflects the initial attraction of the advertisement, indicating whether the user is willing to click on the advertisement, and can screen out advertisements that match the user's interests, quickly excluding low-attraction content. The page residence time is the time the user stays on the advertisement landing page, which can measure the content matching degree of the user to the advertisement landing page. If the residence time is too short, it means that the user quickly jumps out after finding non-target information. If the residence time is too long, it indicates that the user actively browses and reads deeply, and the content highly matches the needs. The repurchase rate is the number of times the user repeats the purchase of the product corresponding to the advertisement, which can reflect the user's long-term recognition and love for the advertisement. Extracting the user's click-through rate, page residence duration, and repurchase rate for each advertisement to evaluate the user preference can more accurately identify the user's interest degree in the advertisement, avoid the increase in click-through rate caused by accidental touch, and filter out invalid clicks by combining the page residence time and repurchase rate.

[0054] It should be further explained that the user preference label is generated based on the potential advertisement. Similarly, by identifying keywords in the potential advertisement and combining multiple keywords into a preference set, the user's short-term behavior preference is reflected, and the preference set is used as the user preference label. Exemplarily, if the keyword identification result for the potential advertisement is "fitness equipment", then the user preference label is {fitness equipment}.

[0055] S5: Advertisement similarity recognition: Based on the user interest label and the user preference label, perform a similarity match, thereby evaluating the advertisement similarity, and sort each advertisement according to the preset sorting order based on the evaluation result.

[0056] In this embodiment, it should be specifically explained that the advertisement similarity evaluation is as follows:

[0057] Generate a low-dimensional vector for the user interest label using pre-trained word vectors and use it as the interest vector. The pre-trained word vectors can specifically be Word2Vec or BERT.

[0058] It should be noted that if there are two or more user interest labels, the generated low-dimensional vectors are weighted and averaged to compound the user interests and form a more complete interest vector.

[0059] Similarly, generate a low-dimensional vector for the user preference label and use it as the preference vector.

[0060] Evaluate the preference similarity between the interest vector and the preference vector through cosine similarity, which is specifically expressed as:

[0061] ,

[0062] where Cs(IT, PT) represents the preference similarity between the interest vector and the preference vector, and vIT and vPT represent the interest vector and the preference vector respectively.

[0063] It should be noted that in personalized advertisement delivery, it is necessary to match the user's abstract interests with specific preferences. In this case, cosine similarity mainly measures the directional difference between two vectors rather than the length. Exemplarily, the vectors of "sports enthusiasts" and "fitness equipment" in the vector space have different lengths, but their vector directions are close, so their similarity degree is relatively high.

[0064] It should be further noted that each advertisement is sorted according to the preference similarity in a preset sorting order. The preset sorting order is in descending order. The advertisement ranked first is the matching advertisement, and the matching advertisement is the advertisement with the highest similarity to the user's preferences.

[0065] S6: Advertisement recommendation and display: Based on the sorting result, display personalized advertisements in real time, trigger advertisement switching through hotkey interaction or gesture operation, and update the labels and sorting according to the user's real-time behavior feedback.

[0066] In this embodiment, it should be specifically noted that the top - ranked matching advertisement is the advertisement that matches the user's preferences. The matching advertisement can generate eye - catching titles, copywriting suggestions, and video scripts through AI, improving the content quality and production efficiency. It can reach users efficiently by unlocking the screen and reducing interference. The top - ranked matching advertisement is preferentially displayed on the splash screen interface to ensure that the user first sees the advertisement that they are most interested in, thereby arousing the user's interest and increasing the user's click - through rate. At the same time, when the user's demand for the matching advertisement decreases, the user can switch to the optimal option after screening and sorting through the interactive hotkey, saving the user's time and energy and improving the matching efficiency between the advertisement and the demand. On the other hand, since forced full - screen advertisements are likely to arouse user disgust, providing an interactive hotkey to switch advertisements gives users control, enabling them to more conveniently view other matching advertisements that they are interested in, thereby increasing the user's interaction depth, increasing the user's stay time and participation.

[0067] As shown in the Figure 2 interest data recognition system of an AI intelligent mobile terminal shown in the appendix, which includes: a system operation database, a system central processor, and a user information terminal, and also includes: an information acquisition module, a user usage evaluation management module, an interest tag acquisition module, a user preference tag acquisition module, an advertisement similarity recognition module, and an advertisement recommendation display module;

[0068] The system operation database includes all data texts of the interest data recognition system and collects information texts output by each module in real - time. The system central processor is used to control the information text instructions output during the entire control process. The user information terminal is an information output device that receives the data analysis system.

[0069] Information acquisition module: Set an advertisement entrance after the mobile terminal starts up, obtain the user's basic information through the advertisement entrance, and allow the user to upload usage evaluations on the advertisement details page;

[0070] User usage evaluation management module: Identify valid evaluations based on the user's usage evaluations and obtain the user's valid scores based on the valid evaluations;

[0071] Interest tag acquisition module: Extract advertisements that meet the user's needs based on the user's valid scores and generate user interest tags through the valid scores of the advertisements;

[0072] User preference tag acquisition module: Obtain the user's behavior data, calculate the user preference coefficient, and generate user preference tags based on the user preference coefficient;

[0073] Advertisement similarity recognition module: Perform similarity matching based on the user interest tags and user preference tags, thereby evaluating the advertisement similarity, and sorting each advertisement according to a preset arrangement order based on the evaluation results;

[0074] Advertising recommendation display module: Perform advertising recommendation display based on the sorting result.

[0075] An interest data recognition medium for an AI intelligent mobile terminal, when executed by a processor, implements an interest data recognition method for an AI intelligent mobile terminal.

[0076] In a specific example, assume that the positive ratings of each advertisement are 4.5, 4.5, 5.0, 5.0 in sequence, and the negative ratings are 2.0, 1.5, 2.5 in sequence. At this time, the median effective rating is 4, the average positive rating is 4.625, and the average negative rating is 2; Use the mobile terminal to collect the click-through rate, page stay time, repurchase rate, and the longest and shortest page stay times of the user for an advertisement, and obtain the click-through rate Cl = 15%, the page stay time Ct = 120s, the repurchase rate Cu = 5%, the longest page stay time is 180s, and the shortest page stay time is 45s respectively;

[0077] Further, the deviation degree of the positive rating corresponding to the positive advertisement is specifically: Po = |4.625 - 4| / 4 = 0.156, and the deviation degree of the negative rating corresponding to the negative advertisement is specifically: Ne = |2 - 4| / 4 = 0.5; The standard value of the stay time is specifically: Ct = (120 - 45) / (180 - 45) = 0.56.

[0078] Then calculate the advertising placement deviation degree, specifically:

[0079] ,

[0080] Substitute the parameters Po = 0.156, Ne = 0.5,

[0081] Calculate ;

[0082] The user preference coefficient is specifically: Dl = 0.4 × Cl + 0.3 × Ct + 0.3 × Cu,

[0083] Substitute the parameters Cl = 0.15, Ct = 0.56, Cu = 0.5, and calculate Dl = 0.4 × 0.15 + 0.3 × 0.56 + 0.3 × 0.5 = 0.24, where 0.4, 0.3, 0.3 are the weights corresponding to the click-through rate, the standard value of the stay time, and the repurchase rate respectively;

[0084] The reference advertisement, relevant advertisement, and potential advertisement are obtained respectively through the advertising deviation and the user preference coefficient. According to the reference advertisement and the relevant advertisement, the keyword recognition results are "sports shoes" and "fitness mat", and the interpretable interest label conversion result of the keyword is "sports enthusiasts". According to the potential advertisement, the keyword recognition result is "fitness equipment", then the user preference label is refined through the preset label library as {fitness equipment};

[0085] Generate a low-dimensional vector as the interest vector through the user interest label, v IT =[0.8, 0.2]. Generate a low-dimensional vector as the preference vector through the user preference label, v PT =[0.6, 0.4];

[0086] The preference similarity between the interest vector and the preference vector is specifically:

[0087] ,

[0088] Substitute the vector v IT =[0.8, 0.2], v PT =[0.6, 0.4],

[0089] Calculate 。

[0090] Finally, store the preference similarity in the system operation database, sort the stored preference similarity in descending order, display personalized advertisements in real time based on the sorting result, and trigger the advertisement to switch to other advertisements that the user is interested in through hotkey interaction or gesture operation.

[0091] Secondly: In the attached drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments are involved. Other structures can refer to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other;

[0092] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An interest data recognition method for an AI intelligent mobile terminal, characterized in that, Including: S1: Information acquisition: Set an advertisement entry after the mobile terminal is powered on. Obtain the user's basic information through the advertisement entry, where the user's basic information is the user's basic personal profile and information about their consumption situation, and allow the user to upload a usage evaluation on the advertisement details page; S2: User usage evaluation management: Identify valid evaluations based on the user's usage evaluations. At the same time, identify the core vocabulary of the valid evaluations and perform a scoring classification, where the scoring classification includes positive scoring and negative scoring; S3: Interest tag acquisition: Based on the user usage evaluation management, extract the proportion of positive scores and classify the advertisement categories, where the advertisement categories include positive advertisements and negative advertisements. Analyze the deviation degree of advertisement placement according to the advertisement categories, and thus generate user interest tags; The process of generating the user interest tags is as follows: Statistically analyze the usage evaluations uploaded by the user, extract the proportion of valid scores, and set the corresponding standard proportion. Extract the positive scores that reach the standard proportion, set the advertisements corresponding to the positive scores that reach the standard proportion as positive advertisements, set the advertisements corresponding to the positive scores that do not reach the standard proportion and negative scores as negative advertisements, and mark the user scores corresponding to the negative advertisements as negative scores; Extract positive scores and negative scores from the valid evaluations of positive advertisements and negative advertisements, calculate the positive scores and negative scores of positive advertisements and negative advertisements, obtain the average positive score and average negative score of each advertisement, and obtain the median valid score by comparing the valid scores of the advertisements; Subtract the average positive score and average negative score of each advertisement from the median valid score, take the absolute value, and then divide by the median valid score to obtain the positive score deviation degree corresponding to the positive advertisement and the negative score deviation degree corresponding to the negative advertisement; Compare the positive score deviation degree corresponding to the positive advertisement and the negative score deviation degree corresponding to the negative advertisement to obtain the advertisement placement deviation degree; Compare the advertisement placement deviation degree of each advertisement with the defined deviation degree. If the advertisement placement deviation degree of the advertisement reaches the defined deviation degree, eliminate the advertisement at this time. If the placement deviation degree of the advertisement does not reach the defined deviation degree, identify the advertisement as a positive advertisement, then set the positive advertisement as a reference advertisement, identify the advertisement as a negative advertisement, then set the negative advertisement as a relevant advertisement; Identify keywords for the reference advertisements and relevant advertisements, and convert the keywords into interpretable user interest tags; S4: User preference tag acquisition: Obtain the user's behavior data through the mobile terminal, and analyze the user preference coefficient corresponding to each advertisement when the user uses it according to the user's behavior data. Judge the user preference degree based on the user preference coefficient, and thus generate user preference tags; S5: Advertisement similarity identification: Perform similarity matching based on the user interest tags and user preference tags, thus evaluate the advertisement similarity, and sort each advertisement according to a preset sorting order according to the evaluation result; S6: Advertisement recommendation display: Based on the sorting result, display personalized advertisements in real time, trigger advertisement switching through hotkey interaction or gesture operations, and update the tags and sorting according to the user's real-time behavior feedback.

2. The interest data recognition method of an AI intelligent mobile terminal according to claim 1, characterized in that: The specific method for identifying valid evaluations is as follows: Segment and clean the evaluation content of each usage evaluation, and perform part-of-speech tagging on the segmented words. Based on the part-of-speech of each segmented word, set adjectives as sentiment descriptor words and nouns as topic words through a predefined sentiment word library. Thus, extract sentiment descriptor words and topic words from the part-of-speech of each segmented word, and combine adjacent sentiment descriptor words and topic words into core vocabulary. If no core vocabulary is identified in the usage evaluation, then discard this usage evaluation.

3. The method for identifying interest data of an AI intelligent mobile terminal according to claim 1, wherein: The specific extraction of the user's valid score is as follows: Obtain the corresponding valid evaluation star rating based on the valid evaluation. The valid evaluation star rating is the star rating given by the user during the usage evaluation, including five stars. Thus, extract the user's valid score. If the user has multiple valid evaluations for this advertisement, then calculate the average value of the valid evaluation star ratings corresponding to the multiple valid evaluations as the user's valid score. Compare the user's valid score with a preset positive score. If the user's valid score is greater than or equal to the preset positive score, then use this valid score as the positive score. If the user's valid score is less than the preset positive score, then use this valid score as the negative score.

4. The method for identifying interest data of an AI intelligent mobile terminal according to claim 1, characterized in that: The process of generating the user's preference tags is as follows: Extract the click-through rate, page stay time, and repurchase rate of the user for each advertisement from the mobile terminal, and integrate them into user behavior data. Extract the longest page stay time and the shortest page stay time based on the user's page stay time, and compare the difference between the page stay time of this advertisement and the shortest page stay time with the difference between the longest page stay time and the shortest page stay time to obtain the stay time standard value. Perform weighted averaging based on the click-through rate, stay time standard value, and repurchase rate corresponding to each advertisement to statistically obtain the user preference coefficient. Compare the user preference coefficient with a defined preference coefficient. If the user preference coefficient is greater than or equal to the defined preference coefficient, then set this advertisement as a potential advertisement. If the user preference coefficient is less than the defined preference coefficient, then discard this advertisement. Identify keywords for potential advertisements, combine multiple keywords into a preference set, and use the preference set as the user preference tag.

5. The method for identifying interest data of an AI intelligent mobile terminal according to claim 1, characterized in that: The specific advertisement similarity assessment is as follows: Use pre-trained word vectors to generate low-dimensional vectors for the user interest tags, and use them as interest vectors. If there are two or more user interest tags, then perform weighted averaging on the generated low-dimensional vectors to compound the user's interests and form a complete interest vector. Similarly, generate low-dimensional vectors for the user preference tags and use them as preference vectors. Evaluate the preference similarity between the interest vector and the preference vector through cosine similarity.

6. The method for identifying interest data of an AI intelligent mobile terminal according to claim 1, wherein: The specific method for sorting the advertisements is as follows: Sort the advertisements according to the preference similarity in a preset sorting order. The preset sorting order is in descending order, and the advertisement ranked first is the matching advertisement.

7. An interest data recognition system for an AI intelligent mobile terminal, according to the method for recognizing interest data of an AI intelligent mobile terminal according to any one of claims 1-6, characterized in that, Including: Information acquisition module: Set an advertisement entrance after the mobile terminal screen is turned on, obtain the user's basic information through the advertisement entrance, and allow the user to upload usage evaluations on the advertisement details page. User usage evaluation management module: Identify valid evaluations based on user usage evaluations, simultaneously identify core vocabulary for valid evaluations, and perform score classification, where score classification includes positive scores and negative scores; Interest tag acquisition module: Extract the proportion of positive scores based on user usage evaluation management, classify advertisement categories, analyze the deviation of advertisement placement according to advertisement categories, and thus generate user interest tags; User preference tag acquisition module: Obtain user behavior data through a mobile terminal, analyze the user preference coefficients corresponding to each advertisement during user usage based on the user behavior data, and judge the user preference degree based on the user preference coefficients, and thus generate user preference tags; Advertisement similarity recognition module: Perform similarity matching based on user interest tags and user preference tags, thereby evaluate advertisement similarity, and sort each advertisement according to a preset arrangement order based on the evaluation result; Advertisement recommendation display module: Perform advertisement recommendation display based on the sorting result.

8. An interest data recognition medium for an AI intelligent mobile terminal, when the interest data recognition medium of the AI intelligent mobile terminal is executed by a processor, implement an interest data recognition method for an AI intelligent mobile terminal as described in any one of claims 1-6.

Images