An artificial intelligence-based advertising method and device for equipment sales

Abstract

The present invention relates to the field of marketing technology, specifically an artificial intelligence-based advertising method and apparatus for equipment sales, comprising the following steps: collecting social media behavior data and purchase records of consumers in the target market, obtaining standardized consumer data through data cleaning and formatting; and performing user portrait analysis based on the standardized consumer data. In the present invention, the collection of user social media behavior data and purchase records, combined with data cleaning and formatting, makes the construction of user portraits more dynamic and accurate. Consumer segmentation not only relies on basic demographic information, but also introduces purchase frequency, interest tags, and historical interactive behaviors to form a more hierarchical user classification. The adjustment of advertising content is not limited to static text optimization, but also dynamically adjusts visual elements, language style, and information presentation methods based on the results of sentiment analysis to make the content more in line with the preferences of different consumer groups.

Description

Technical Field

[0001] The present invention relates to the field of marketing technology, and in particular to an artificial intelligence-based advertising method and device for equipment sales. Background Art

[0002] The field of marketing technology encompasses the promotion, sales, and distribution of products and services from manufacturer to consumer, encompassing market research, consumer behavior analysis, brand management, marketing strategy development, advertising, and sales channel optimization. Device sales advertising is a promotional strategy within the marketing technology field that aims to increase market awareness, customer interest, and sales conversion rates for device products through the design and delivery of precise advertising content.

[0003] Existing consumer segmentation technologies rely on traditional demographic information to segment users, ignoring individual behavioral characteristics and dynamic interest changes. This makes ad targeting susceptible to the limitations of static data and makes it difficult to accurately match user needs. Ad content adjustment primarily relies on fixed templates, failing to incorporate user emotional feedback for personalized optimization. This lack of targeted information presentation leads to significant homogeneity in ad content, reducing user engagement and click-through rates. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an advertising method and device for equipment sales based on artificial intelligence.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an artificial intelligence-based advertising method for device sales, comprising the following steps:

[0006] Collect social media behavior data and purchase records of consumers in the target market, and obtain standardized consumer data through data cleaning and formatting. Based on this standardized consumer data, conduct user profile analysis, including age, preferences, and purchasing power, to generate consumer segmentation results.

[0007] Based on the consumer segmentation results, sentiment analysis is performed on the consumer groups to identify emotional reactions and preferences to the advertising content and obtain sentiment preference results; advertising copy and visual elements are adjusted according to the sentiment preference results to generate customized advertising content;

[0008] Using the customized advertising content, testing is performed on the platform, user interaction data and feedback are recorded, and advertising effectiveness evaluation results are obtained; based on the advertising effectiveness evaluation results, performance differences on different platforms are analyzed, advertising delivery strategies are optimized, and optimized advertising deployment plans are generated;

[0009] According to the optimized advertisement deployment plan, advertisements are published, and real-time data streams are monitored at the same time. Advertisement content and delivery time are adjusted according to the real-time data streams to obtain delivery results.

[0010] Preferably, the steps for obtaining the standardized consumer data are:

[0011] Collect social media behavior data and purchase records of consumers in the target market, remove duplicate data, correct data format errors, and obtain a preliminarily processed data set;

[0012] Based on the preliminarily processed data set, missing values, outliers and data deviations are identified and processed, and data interpolation and outlier correction are applied to obtain standardized consumer data.

[0013] Preferably, the steps for obtaining the consumer segmentation results are:

[0014] Extracting the consumer's age, purchase frequency, and preferred category from the standardized consumer data to obtain a feature extraction result;

[0015] Based on the feature extraction results, the purchasing power index of each consumer is calculated using the following formula:

[0016]

[0017] in, Representing consumers Purchasing power index, Representing consumers The age value, Represents the median age of all consumers. Representing consumers Purchase frequency, Representing consumers Types of preferences, Represents the sum of the purchase frequency values ​​of all consumers in the target market. Representing consumers Total purchase amount;

[0018] Based on the purchasing power index, combined with consumers' purchase frequency, preferred categories and age range, consumers are categorized and a consumer segmentation result is generated.

[0019] Preferably, the steps for obtaining the emotion preference result are:

[0020] Extracting advertising interaction data for each group from the consumer segmentation results, including the number of likes, number of comments, sharing frequency, and ad viewing time, and performing data screening and classification to obtain consumer group interaction characteristics;

[0021] Based on the interaction characteristics of the consumer groups, the emotional response score of each group to the advertising content is calculated using the following formula:

[0022]

[0023] in, Representing a group The emotional response score, Representative groups Number of likes, Representative groups Number of comments, Representative groups Frequency of ad sharing, Representative groups ad viewing time;

[0024] Based on the emotional response scores and combined with the advertising interaction patterns of the consumer groups, the emotional trends of each consumer group towards the advertising content are analyzed to obtain emotional preference results.

[0025] Preferably, the steps for obtaining the customized advertising content are:

[0026] Based on the emotional preference results, extract the preferences of different consumer groups for advertising text and visual design, and generate advertising copy features and visual element features;

[0027] Based on the characteristics of the advertising copy and visual elements, screen the advertising text content and visual materials that meet the emotional preferences of the consumer group, adjust the expression of text sentences, the presentation of advertising themes and the structure of information transmission, and generate advertising copy adjustment plans and visual element adjustment plans;

[0028] Based on the advertising copy adjustment plan and the visual element adjustment plan, the optimized copy content and visual materials are integrated, the matching degree between the advertising content and the characteristics of the consumer group is adjusted, and customized advertising content is generated.

[0029] Preferably, the steps for obtaining the advertising effect evaluation result are:

[0030] Using the customized advertising content, conduct advertising delivery tests on the platform, set advertising delivery strategies, including release time, display frequency, and target audience, collect user interaction behavior data, and generate a user interaction data set;

[0031] Based on the user interaction dataset, the advertisement suitability score is calculated using the following formula:

[0032]

[0033] in, Indicates advertising The fitness score of Representative Advertising Number of clicks, Representative Advertising The residence time, Representative Advertising Number of shares, Representative Advertising Number of comments, Representative Advertising Number of impressions;

[0034] Based on the advertisement suitability score, the performance of the advertisement on the platform is analyzed, and the user interaction effects of each advertisement are compared to obtain an advertisement effect evaluation result.

[0035] Preferably, the steps for obtaining the optimized advertisement deployment plan are:

[0036] Based on the advertising effectiveness evaluation results, calculate the performance differences of advertising on various platforms, analyze user behavior patterns on different platforms, compare the effectiveness of advertising content among different audience groups, and generate advertising strategy optimization plans;

[0037] Based on the advertising delivery strategy optimization plan, adjust the advertising delivery time, adjust the display method of advertising content on various platforms, optimize the adaptation strategy of advertising materials, adjust the budget allocation ratio, and form an optimized advertising deployment plan.

[0038] Preferably, the steps for obtaining the delivery results are:

[0039] Delivering advertisements according to the optimized advertisement deployment plan, monitoring user behavior data after advertisement exposure, and generating real-time advertisement feedback data;

[0040] Based on the real-time advertising feedback data, the advertising delivery result is calculated using the expression:

[0041]

[0042] in, Representative Advertising The advertising results, Representative Advertising In time User page interaction intensity within Representative Advertising Depth of content browsing, Representative Advertising Page bounce rate, Representative Advertising The average length of stay in each time period, Representative Advertising The number of repeated visits by users, T represents the total duration after the ad is delivered;

[0043] Based on the advertisement delivery results, the user response to the advertisement on each platform and time period is analyzed to obtain the delivery results.

[0044] The present invention provides an advertising device, comprising: a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory, so that the advertising device executes the above-mentioned artificial intelligence-based advertising method for device sales.

[0045] Compared with the prior art, the advantages and positive effects of the present invention are:

[0046] In the present invention, by collecting user social media behavior data and purchase records, combined with data cleaning and formatting processing, the construction of user portraits is made more dynamic and accurate. Consumer segmentation not only relies on basic demographic information, but also introduces purchase frequency, interest tags and historical interactive behavior to form a more hierarchical user classification. The adjustment of advertising content is not limited to static text optimization, but also dynamically adjusts visual elements, language style and information presentation methods based on the results of sentiment analysis to make the content more in line with the preferences of different consumer groups. During the advertising delivery stage, a platform testing mechanism is adopted to record user interaction data, and combined with feedback information, a multi-dimensional advertising effect evaluation system is established to enable the delivery strategy to have dynamic adaptability. For advertising deployment, the real-time data flow monitoring mechanism enables the delivery plan to have adaptive adjustment capabilities. According to market feedback, the display method, release time and interactive guidance strategy are adjusted to improve the response efficiency of advertising content. The full-process data closed loop makes the advertising dissemination path more visual and optimizable, reduces resource waste, and improves advertising coverage and conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0049] See also Figure 1 The present invention provides a technical solution, an artificial intelligence-based device sales advertising method, comprising the following steps:

[0050] Collect social media behavior data and purchase records of consumers in the target market, and obtain standardized consumer data through data cleaning and formatting. Based on standardized consumer data, conduct user profile analysis, including age, preferences, and purchasing power, to generate consumer segmentation results.

[0051] Based on the consumer segmentation results, sentiment analysis is performed on consumer groups to identify emotional reactions and preferences to advertising content and obtain emotional preference results. Ad copy and visual elements are adjusted based on the emotional preference results to generate customized advertising content.

[0052] Utilize customized advertising content to conduct tests on the platform, record user interaction data and feedback, and obtain advertising effectiveness evaluation results. Based on the advertising effectiveness evaluation results, analyze performance differences on different platforms, optimize advertising delivery strategies, and generate optimized advertising deployment plans.

[0053] According to the optimized advertising deployment plan, advertisements are published, and real-time data streams are monitored at the same time. The advertising content and delivery time are adjusted according to the real-time data stream to obtain the delivery results.

[0054] The steps to obtain standardized consumer data are:

[0055] Collect social media behavior data and purchase records of consumers in the target market, remove duplicate data, correct data format errors, and obtain a preliminarily processed data set;

[0056] Based on the preliminarily processed data set, missing values, outliers and data deviations are identified and processed, and data interpolation and outlier correction are applied to obtain standardized consumer data.

[0057] Specifically, based on the previously acquired consumer social media behavior data and purchase records, we first check whether there are consecutive records of the same user according to the pre-set duplication judgment criteria. For example, when the user number is the same and the record time difference is within If the time is less than 1 second, it is considered repeated. The second threshold is obtained by adding the average value of the repeated samples collected in the past Seconds of experience redundancy, through this method, all data entries are traversed and compared, records that meet the above duplication conditions are eliminated, and then the format of text fields and numerical fields is checked to see if they meet the agreed specifications, such as whether the date format meets The format or numeric field contains non-numeric characters. If confusion is found (for example, the month and day are swapped or the time zone information is wrong), the system will compare them one by one according to the pre-prepared time calibration comparison table, which records common date and time zone combinations and their matching ratios. When the matching ratio is greater than If the format is correctable, the string will be split according to the manually set segmentation rules and then compared. If a symbol error is detected in the numeric field (such as an extra "," or "."), it will be directly removed or replaced with an integer. The judgment threshold for the integer form here is based on the statistical distribution of common currency amounts in the target market. Assume that The price value falls within to If a value exceeds this range, it will be regarded as a suspected format abnormality. Then, combined with the user's purchase volume or transaction type, it will be determined whether it needs to be corrected or eliminated. Finally, all records that have undergone repeated checks and format corrections will be collected together to obtain a preliminarily processed data set.

[0058] Based on the preliminarily processed data set obtained above, we first scan each item to identify whether there are missing values. When it is found that a field is empty or a symbol is used to fill in the missing value, we interpolate it with the historical average value with the same meaning as the field. The historical average value is a comprehensive calculation of the median and mean of the same type of field data selected from the past three months. For example, for a single amount field, the mean value can be calculated within the range of 0 to 10,000 yuan and the compensation value can be calculated based on the transaction frequency to ensure that the interpolated value has a reasonable distribution. Then we continue to detect outliers and data deviations, and compare each record with the preset upper and lower limits. For example, if the age is limited to to If the range is exceeded, the quartile method is used to check the difference from the normal distribution. If the range exceeds a certain multiple of the sum of the distances between the third quartile and the first quartile (the multiple is evaluated by the common extreme value positions in the training set, and the experience is set to ) is temporarily identified as an abnormal record, and then combined with the number of purchases or preference types of the user to determine whether to retain it. If it is determined to be abnormal, it can be further corrected using weighted interpolation, such as correcting based on the mean and variance of the field for the same group of people. The interpolation weight threshold will be calculated during the correction. for ,Should The value comes from the prediction deviation test results of the model on 10,000 samples. If deviation still occurs after interpolation, the record will be marked as pending verification. All entries that have completed missing and abnormal processing are summarized after this process is completed, and the data interpolation outlier correction operation is completed immediately to obtain standardized consumer data.

[0059] The steps to obtain consumer segmentation results are:

[0060] Extract the consumer's age, purchase frequency, and preference categories from the standardized consumer data to obtain feature extraction results;

[0061] Based on the feature extraction results, the purchasing power index of each consumer is calculated using the following formula:

[0062]

[0063] in, Representing consumers Purchasing power index, Representing consumers The age value, Represents the median age of all consumers. Representing consumers Purchase frequency, Representing consumers Types of preferences, Represents the sum of the purchase frequency values ​​of all consumers in the target market. Representing consumers Total purchase amount;

[0064] Based on the purchasing power index, combined with consumers' purchase frequency, preferred categories and age range, consumers are categorized and consumer segmentation results are generated.

[0065] Specifically, based on the age, purchase frequency and preference category information marked in the standardized consumer data obtained earlier, each record is concatenated according to the consumer's unique number, and the corresponding age field, purchase frequency value and preference category mark are read. In this process, a classification comparison table is first referred to to determine the attribution of the preference category. The comparison table is mapped by common product types and attributes in the target market. When the product label of a record matches a category in the comparison table by more than 80%, the consumer's transaction is classified into this category. When the matching degree is less than 80%, it is further compared according to the lower-level classification catalog. After confirming the preference category, the transaction is summarized. The total number of purchases is recorded and the age field is recorded. The age field is compared with the pre-set valid range of 0 to 120 years old. When a consumer's age is found to be less than 0 or greater than 120, a record comparison operation is performed. By looking for adjacent records under the same number and global statistical data, it is determined whether it is an input deviation. If the input deviation is confirmed, the record is eliminated. The purchase frequency value is also limited to between 1 and 1000 times and checked. Exceeding this range indicates that the corresponding record has an extreme anomaly. The entire transaction data under the number can be queried again based on its frequency of occurrence for verification. Finally, all records that meet the conditions are packaged and summarized to obtain the feature extraction results.

[0066] The benefit of the formula is that it combines the age difference item and the purchasing behavior item in the numerator and introduces the square root of the total purchase frequency and consumption amount in the denominator to comprehensively measure the performance of consumers in different dimensions. This allows consumers whose age deviates further from the median age of market entities and whose purchasing behavior is more active to present higher or lower values ​​in the purchasing power index. This numerical difference helps to accurately distinguish different types of consumer groups in subsequent segmentation links.

[0067] The parameters are obtained by the consumer The age value is directly read from the standardized consumer data obtained above. The age value is obtained through ID card registration information or user-initiated information, and is deduplicated and processed during collection. For some parallel inputs (for example, when the same consumer fills in conflicting ages on different occasions), the actual registered valid ID records are retrieved for comparison. If the age field is found to be outside the range of 0 to 120 years old, it will be regarded as an input anomaly, and then corrected or eliminated by comparing with adjacent records. In a test of 500,000 registered users, the age field values ​​of about 2% of the population deviated from the reasonable range, and finally a stable range was obtained by adopting the corresponding correction method. value.

[0068] The steps to obtain the parameter are as follows: it is the median of all consumer age values. It is necessary to sort the complete user age list and take the median value. Assuming that a total of N valid age records are collected, when N is an odd number, the age value at the middle position after sorting is taken. When N is an even number, the average of the values ​​at the middle two positions is taken. For example, in the scenario of collecting age data of 600,000 consumers, after all ages are sorted, the ages at the 300,000th and 300,001st positions are 32 and 33 respectively. Then = =32.5.

[0069] The parameters are obtained as follows: The purchase frequency within a specified period can be obtained by querying the historical transaction database and summarizing all the orders placed by the consumer within the period. For example, if a consumer completes 45 orders within a year, then =45. For further analysis, we can combine the order timestamps. For multiple orders placed on the same day, we count each order independently. By counting the order data of 500,000 users over a year, we accumulate the number of transactions for each user within this period and form a purchase frequency array. The purchase frequency value of each user serves as the specific value of this parameter.

[0070] The parameters are obtained by the consumer To determine the preference types, first classify the types of goods involved in multiple orders of consumers, and then map the types of goods to numerical values. For example, assign the value 2 to the electronic products category, 3 to the home life category, and 5 to the catering food category. If a consumer purchases the same category of goods multiple times within a specified period, the category with the highest frequency is taken as the main preference of the consumer at the current stage, and the numerical value is used as the value of the category. For example, a consumer places 50 orders in a year, 30 of which are electronic products, 12 are food and beverages, and 8 are other categories. =2.

[0071] The steps to obtain the parameter are as follows: it represents the sum of the purchase frequency values ​​of all consumers in the target market. First, select the analysis period, then summarize the purchase frequency of each consumer in the period one by one, and then add them up after the summary is completed. For example, if 300,000 consumers are counted in a one-year period, and the total number of orders from all consumers is 6,000,000, then =6000000.

[0072] The parameters are obtained as follows: The total purchase amount of the consumer is first retrieved from the transaction database. The payment amount of all orders within a specified period is then added up to get a total value. For example, in a consumer's order data, 30 orders have a total payment amount of 3,000 yuan, 15 orders have a total payment amount of 1,800 yuan, and the remaining 5 orders have a total payment amount of 450 yuan. =3000+1800+450=5250.

[0073] Calculation process:

[0074] Make a consumer The relevant parameter values ​​are , , , , , , first calculate = =7, and then = , then ask =30+2=32, the numerator is , the denominator is calculated first = , and compare it with Add together and get =500059.1608, from which we get:

[0075]

[0076] The results show that in this example, the purchasing power index value of this consumer is about 0.0001693. When the same calculation is performed on all consumers under the same market dimension, different intervals can be set for the obtained purchasing power index sequence for segmented comparison. When the value is significantly higher than the overall mean, it means that there is a significant difference between the age and the median and the purchasing behavior is relatively active. When the value is too low, it means that the consumer is not outstanding in terms of activity or amount of contribution. Therefore, in the subsequent steps, consumers can be divided into different categories based on these values.

[0077] Based on the purchasing power index obtained above, combined with the consumer's purchase frequency and corresponding preference categories, as well as the age range label established by the age field, a classification judgment criterion needs to be established before starting to perform category attribution judgment. The criterion usually sets several intervals, each of which contains a purchasing power index range and one or more preference category values. First, the consumer's purchasing power index is positioned within the known interval group. For example, the purchasing power index is divided into several segments with a step size of 0.0001. When a consumer's value is between 0.0001 and 0.0003, it can be regarded as the middle segment, and the purchase frequency is between 2 When the purchasing power index is between 0 and 40 and the preference category is equal to 2 or 3, it is judged as one group. When the purchasing power index is close to 0.001 or higher, it is classified into a more active group, and when the purchasing power index is lower than 0.00005, it is included in another group. Combined with the distinction of age ranges, multiple different population dimensions can be distinguished. After the consumer's category is determined, a segmentation label is generated for each consumer. Once a parameter update or new changes in consumer behavior data are found, the classification process can be re-executed and the changing trajectory of the consumer's category in different time periods can be compared to finally obtain the consumer segmentation results.

[0078] The steps to obtain the emotional preference results are:

[0079] Extract advertising interaction data for each group from the consumer segmentation results, including the number of likes, comments, sharing frequency, and ad viewing time. Filter and classify the data to obtain consumer group interaction characteristics.

[0080] Based on the interaction characteristics of consumer groups, the emotional response score of each group to the advertising content is calculated using the following formula:

[0081]

[0082] in, Representing a group The emotional response score, Representative groups Number of likes, Representative groups Number of comments, Representative groups Frequency of ad sharing, Representative groups ad viewing time;

[0083] Based on the emotional response score and combined with the advertising interaction patterns of consumer groups, the emotional trends of each consumer group towards advertising content are analyzed to obtain the emotional preference results.

[0084] Specifically, based on the group information indicated in the consumer segmentation results obtained above, cross-search each group record one by one, read the number of likes field from it and judge whether it is too high or too low according to a pre-set value range. For example, the number of likes is compared with the interval of 0 to 500 times and the group whose number of likes obviously exceeds the interval is marked as a special value. The same interval comparison method is also established for the number of comments. The number of comments is compared with the range of 0 to 1000. Any group with more than 1000 comments is regarded as an abnormal concentration of comments. Then the advertising sharing frequency of the group is retrieved and compared with the threshold range between 1 and 50 times. When there is a value higher than 50 times, it is considered an abnormal concentration. The actual scenarios are traced back based on the sharing records of specific members in the group, and the advertising viewing time is compared with the interval of 3 seconds to 600 seconds. During the inspection process, if there is missing data in the number of likes, comments, sharing frequency or viewing time, it will be marked in time. Then all data items within the normal range will be classified into different categories. For example, the group with the number of likes and comments in the middle range will be classified into one category, and the group with a significantly high sharing frequency but a smaller viewing time will be divided into another category. After the classification is completed, the outlier group will be separated and managed from the other groups. Finally, the performance combination of each group in likes, comments, sharing and viewing time will be summarized to obtain the interactive characteristics of the consumer group.

[0085] The benefit of the formula lies in that by adding the square of the number of likes and the number of comments into the numerator, introducing the square root of the ad sharing frequency into the denominator and adding a benchmark value, and combining it with the logarithmic operation of the viewing time, a single score can comprehensively reflect whether the like behavior is relatively concentrated, whether the comment interaction is active, the intensity of sharing activities and the distribution of viewing time, so that the core interaction characteristics of different groups can be quantified and measured in the same numerical framework.

[0086] The steps to obtain the parameters are as follows, which represents the group The total number of likes in a given delivery period, first from the group Read the like behavior entries from all internal individual user records, use the user's unique identification information as the index to accumulate the number of likes in the complete interactive data one by one. The record objects include various dimensions such as text content likes, image likes and short video likes. When the statistics are completed, these accumulated likes are added up to form In a monitoring of 20,000 ad interaction samples, the likes items are updated daily to confirm the total number of likes for each group within the 7-day delivery cycle. For example, if the statistics of the user group numbered 3 show that this group has generated 260 likes within 7 days, then =260.

[0087] The steps to obtain the parameters are as follows, which represents the group The number of comments received needs to be retrieved from the comment database and merged based on the comment behavior of all users in the group. First, lock the group The identifier corresponding to the user is found and the comment release record with the same identifier in the specified advertising environment is found. The number of comments in all relevant records is summed up to get For example, for users with group number 3, the monitoring range is set to one month. If the total number of comments in one month reaches 120, then =120.

[0088] The steps to obtain the parameters are as follows, which represents the group The frequency of sharing during the advertising period, first count the group when obtaining The number of times all users share the target ad is calculated, and then the number of shares is divided by the number of monitoring days to get an average daily share value. The average daily share value is then multiplied by the total number of days the ad is exposed in this group to form the sharing frequency. In a 10-day trial, if there are 90 shares under group number 3, the average number of shares per day is 9. Since the ad is run for 10 days, =9×10=90.

[0089] The steps to obtain the parameters are as follows, which represents the group The viewing time of the advertising content is recorded by calculating the time when each user opens the advertising content and the time when the user closes the advertising content to obtain the number of seconds for a single viewing. Then, the viewing seconds of all users in the group are accumulated within a given period to obtain the total viewing time. For example, in a 14-day advertising run, the total viewing time of all users in group number 3 is 5400 seconds. =5400.

[0090] Calculation process:

[0091] make , , , , first calculate the molecular part = =900+12=912, then the denominator = = ,get:

[0092]

[0093] Then calculate = = , add the two together, that is:

[0094]

[0095] This result shows that in the current example, the group The emotional response score of different groups is about 349.9563. When comparing the value with this value, groups with higher values ​​tend to be more active in likes, comments, sharing and viewing time, while lower values ​​suggest that the group has less investment in these interactive dimensions. In the actual delivery environment, different Adopt corresponding promotion strategies or content optimization ideas in each interval.

[0096] After obtaining the emotional response scores of each group, a specific comparison is carried out on each group based on the pre-organized advertising interaction pattern classification list. First, the strength level of the click behavior and stay behavior of each group is read, and the distribution of likes, comments and shares made by the group in the past period is used as a reference. When it is found that the emotional response scores of some groups are between 300 and 400 and the number of comments has continued to increase in the recent period, this group will be classified as an emotional trend with a focus on positive evaluation. If the emotional response scores of some groups are between 100 and 200 and the corresponding number of likes increases slightly but the number of comments remains in a stable range , it is regarded as a neutral emotional trend. In addition, when the emotional response score is lower than 50, it will be confirmed in combination with the viewing time and sharing frequency characteristics of the group. Once the viewing time and sharing frequency are at a low level between 5 and 10 times, it will be classified as a low-investment group. At the same time, for emotional response scores above 600, by comparing their overall interaction indicators, such as more than 800 likes or more than 500 comments, it shows that this group shows a stronger advertising relevance. After completing the cross-statistics of all emotional response scores and interaction patterns, the emotional type and tendency of each group towards the advertising content during this stage are recorded one by one to obtain the emotional preference results.

[0097] The steps to obtain customized advertising content are:

[0098] Based on the emotional preference results, we extract the preferences of different consumer groups for advertising text and visual design, and generate advertising copy features and visual element features;

[0099] Based on the characteristics of advertising copy and visual elements, screen advertising text content and visual materials that meet the emotional preferences of consumer groups, adjust the expression of text sentences, the presentation of advertising themes and the structure of information transmission, and generate advertising copy adjustment plans and visual element adjustment plans;

[0100] Based on the advertising copy adjustment plan and the visual element adjustment plan, the optimized copy content and visual materials are integrated, the matching degree between the advertising content and the characteristics of the consumer group is adjusted, and customized advertising content is generated.

[0101] Specifically, according to the emotional preference results obtained above, after reading the corresponding characteristics of different consumer groups marked therein, first compare the reference data of the group in terms of text reading habits and color acceptance, and check one by one whether there is a resistance value to specific modifiers or a record of the intensity of preference for certain visual elements in the group by retrieving the previously established association comparison table. If it is found that the resistance value is greater than 3, it means that the corresponding vocabulary will trigger a more obvious rejection emotion within the group. The threshold of 3 comes from the negative reaction rate when investigating 1,000 groups using words with different emotional colors and conducting statistics. At the same time, the items with an acceptance of bright colors greater than 60% in each group are screened and rearranged. The 60% empirical threshold is obtained by a one-to-one mapping of past advertising data. When the mapping result shows that the majority of people have a certain color exposure of more than 60%, it is considered to have a clear acceptance tendency, and then these analysis information are analyzed. For example, the group that prefers a more straightforward expression style and a stronger main color tone can be grouped into one tendency group, and the group that prefers a softer narrative logic and a more neutral color scheme can be grouped into another tendency group. After the grouping is completed, the extracted modifying words, adjective phrases, color matching schemes, etc. can be further compared across groups to confirm the overlapping parts. Once certain words or visual design forms are found to be obviously attractive to multiple groups, they will be marked as common elements and recorded in the database. When all the more important words, colors, background graphics and other elements are marked, they can be matched with the specific tendencies of each group. If a group shows a greater acceptance of warm-toned colors and shows stable positive feedback on promotional copy, the priority combination of this type of copy and visual composition will be stated in the mark. Finally, the tendencies of all groups are integrated to form corresponding advertising text selection standards and visual configuration instructions, and generate advertising copy characteristics and visual element characteristics.

[0102] Based on the characteristics of advertising copy and visual elements, we first select a number of text sentences that are highly consistent with the emotional preferences of the consumer group and establish an internal sentence index table. The entries in the index table include specific modifiers, paragraph order, and potential image style tags. When encountering expressions with overly inflammatory words or expressions that conflict with the previous threshold judgment, we immediately determine whether to remove them based on the emotional rejection level calculated previously. If the rejection level exceeds 1, it means that there are inappropriate elements and they need to be removed or replaced in the sentence. The rejection level is the ratio of the number of negative reviews to the total number of readings after collecting more than 200 negative reviews. Through this comparison, we get an average proportion, and those exceeding 1 are considered inappropriate for use. After the sorting is completed, the presentation methods of the advertising themes should also be compared one by one. For example, for the same theme, mixed text and pictures or short videos can be used. By means of snippet preview and other methods, the reading time, click-through rate and interaction rate of each consumer group are compared in parallel. When it is found that a certain method can maintain a high click rate in most groups but the number of comments is less than 3 for a long time, it is marked as a presentation method with strong appeal but insufficient interactive depth. In the information transmission structure part, it is necessary to detect whether there are lengthy text segments that cause reading to exit midway. Statistics and comparison are made to see whether the proportion of samples with a page stay time of more than 10 seconds remains above 70%. If it is less than 70%, it means that the information hierarchy needs to be shortened or the order needs to be adjusted. When adjusting, the key text is placed in front first and distinguished by color or icons to prevent visual crowding. Finally, the text content verified by the comparison table is matched with the image material respectively, and the text-visual comparison position is listed for each matching scene to generate the advertising copy adjustment plan and the visual element adjustment plan.

[0103] Based on the advertising copy adjustment plan and the visual element adjustment plan, we first read the appropriate sentence structure from the text sentence index table and correspondingly splice it with the marked visual materials. We package those color presets that are known to be consistent with the group characteristics into the same theme structure. At the same time, we assign a certain order to the key text paragraphs in the information transmission structure so that users can see the core content as soon as they enter the page. When integrating, we need to check whether the contrast between the color scheme and the text meets the readability requirements set earlier. The readability requirements are derived from measuring the text reading accuracy of 150 users and considering color schemes with a contrast ratio greater than 4.5:1 as The optimal solution is to reselect other materials or add a color block at the bottom of the image if a background conflict is found between the existing illustrations and the text overlay area. After all parts are matched, write the text presentation position, image proportion, theme combination and other information into the configuration list item by item, and use visual inspection to determine the display size and layout distribution ratio. Any differences in font size or layout in a multi-terminal environment are also recorded one by one. When the final text combination and visual materials are in line with the preferences of each group and no reading difficulties or content ambiguity occur in the trial preview phase, it is confirmed that this set of solutions can be used for multi-platform delivery to generate customized advertising content.

[0104] The steps to obtain the advertising effectiveness evaluation results are as follows:

[0105] Conduct advertising testing on the platform using customized advertising content, set advertising strategies, including release time, display frequency, and target audience, collect user interaction behavior data, and generate user interaction data sets;

[0106] Based on the user interaction dataset, the ad suitability score is calculated using the following formula:

[0107]

[0108] in, Indicates advertising The fitness score of Representative Advertising Number of clicks, Representative Advertising The residence time, Representative Advertising Number of shares, Representative Advertising Number of comments, Representative Advertising Number of impressions;

[0109] Based on the ad suitability score, the performance of ads on the platform is analyzed, the user interaction effects of each ad are compared, and the ad effectiveness evaluation results are obtained.

[0110] Specifically, based on the customized advertising content obtained earlier, a basic list of delivery strategies should be established before executing the delivery test. The delivery time should be limited to the twelve-hour range from 9:00 to 21:00 every day and segmented at 30-minute intervals. The display frequency should be limited to a maximum of 3 times in each time period and the page stay after each exposure should be recorded. The delivery audience should be targeted at registered users aged 18 to 45 and re-screened based on interest tags. When delivering different advertising content, the overlap between the delivery period and the corresponding population should be checked one by one to avoid repeated sending of more than 5 overlapping exposures. After the delivery begins, the number of clicks, comment items and possible sharing behaviors should be monitored in real time. The delivery progress in each time period is statistically compared. If the number of clicks is greater than 100 or the number of comments exceeds 10, the time period is marked as a high-activity segment. The empirical thresholds of 100 times and 10 comments are derived from the aggregated average of the previous 10,000 delivery data and are corrected based on the results of a field observation. During the statistical process, all relevant user browsing time information is aggregated and associated with their interactive actions. The interactive behavior data is recorded line by line to form a user interaction data set. After the delivery cycle ends, the user's viewing completion rate and page bounce distribution are uniformly checked to supplement the complete interactive indicator items. Finally, all behavioral statistical items are merged to form the final data to generate a user interaction data set.

[0111] The usefulness of the formula lies in combining the number of clicks and dwell time in the numerator, taking the logarithm of the number of shares to prevent extreme sharing values ​​from being overly amplified, and independently adding the ratio of the number of comments to the number of impressions, so that different dimensions of interactive behavior can be balanced and measured within the same scoring system.

[0112] The steps to obtain the parameter are as follows, which represents the advertisement The number of clicks during the delivery period is calculated by first recording the total number of clicks on the ad in each time period, and then adding up the number of clicks in each time period at the end of the day to form the total number of clicks for the day. The total number of clicks accumulated at the end of the delivery period is obtained. For example, if the daily clicks on an ad are 50, 60, 55, 40, 75, 65, and 80 times during a 7-day delivery period, then =50+60+55+40+75+65+80=425.

[0113] The steps to obtain the parameter are as follows, which represents the advertisement The dwell time is obtained by collecting the difference between the start and end browsing time of the user on the ad page and accumulating them. At the same time, a lower limit benchmark value of 3 seconds is set. When the dwell time is less than 3 seconds, it is still regarded as a browse but counted in the "very short stay" category. This 3-second benchmark value is obtained by taking the median of the average reading start time of 1,000 users on the graphic page. When the cumulative dwell time of the ad reaches 30,000 seconds within 7 days, it is =30000.

[0114] The steps to obtain the parameter are as follows, which represents the advertisement First, read all the sharing behavior records of users during the ad display period and lock the unique identifier of the ad. If the ad is shared 200 times in total during the 7-day run, then =200.

[0115] The steps to obtain the parameter are as follows, which represents the advertisement The number of comments is obtained by summing up the number of comments submitted by users under the ad page, adding up the number of comments per day. It is also necessary to identify the comments repeatedly. If the same user posts the same comment in a very short interval, it will be marked as one. For example, if the cumulative number of comments for an ad in this period is 90, then =90.

[0116] The steps to obtain the parameter are as follows, which represents the advertisement By recording the exposure frequency of the advertisement in each time period, the multiple refreshes of the same user in the same time period are regarded as one display. If the advertisement is displayed 2000 times in total by the system after the statistical period ends, =2000.

[0117] Calculation process:

[0118] make , , , , , first find = , and compare it with Add , then =200+1=201 Take the natural logarithm , we can get the first term by dividing the numerator by the denominator:

[0119]

[0120] Then calculate the second term = = , after adding the two ;

[0121] The results show that when When the number is above 100, the ad is more active in terms of click performance, dwell time, sharing, and comments. By summarizing and sorting the values, we can compare the interaction strength of different ads. A higher value means more attention it receives at the user level, while a value that is too low means its appeal or interaction is limited. These quantitative data can help complete subsequent differentiated analysis and optimized deployment.

[0122] After calculating the ad fit score, we first collect a list of scores for each ad and place them alongside the corresponding delivery parameters. We then retrieve key engagement metrics for each ad during its delivery cycle, such as click-through rate, dwell time, and comment trends. We compare these metrics to other ads delivered during the same time period to examine the gap between high- and low-activity segments. Ads with scores over 200 are re-checked based on metrics like over 500 clicks and over 100 comments to confirm whether they have shown significant growth in user engagement and subsequent sharing intentions. Ads with scores below 50 are marked as underengaged and their impression-to-click ratio is checked to see if it is less than 1:100. If this ratio is less than 1:100, it indicates that the ad's impressions far outweigh its engagement response, allowing for further tracing of comment and share records. Once we confirm that there are no duplicates or outliers in the score sequence, we number all ads from highest to lowest score and perform data visualization on each ad. Finally, we compile a comprehensive comparison chart of the platform's ad engagement levels, listing all scores and core metrics and annotating the difference ranges to provide an evaluation of ad effectiveness.

[0123] The steps to obtain the optimized advertising deployment plan are:

[0124] Based on the results of advertising effectiveness evaluation, calculate the performance differences of advertising on various platforms, analyze user behavior patterns on different platforms, compare the effectiveness of advertising content among different audience groups, and generate advertising strategy optimization plans;

[0125] Based on the advertising strategy optimization plan, adjust the advertising delivery time, adjust the display method of advertising content on various platforms, optimize the adaptation strategy of advertising materials, adjust the budget allocation ratio, and form an optimized advertising deployment plan.

[0126] Specifically, based on the results of advertising effect evaluation, all obtained interaction indicators are first classified by platform, and the exposure time, number of clicks and number of comments of the same advertisement on different platforms are split according to the audience type. The user behavior records of each platform are read, and each record is compared with the pre-established effective range. For example, the number of daily active users is compared with the range of 1,000 to 50,000, the average stay time per person is compared with the range of 10 seconds to 300 seconds, and the interaction frequency is compared with the range of 0 to 10 times. If it is found to be out of the range, the record is marked as an outlier and the data is checked. After confirmation, it is summarized to form a platform difference comparison item. The audience of each platform is further decomposed based on the comment preference and sharing tendency. For example, if the number of comments is greater than 100 but the number of likes is less than 20 on some social platforms, it is judged as a pattern with active comments but low click-through conversion. In another part of the information platform, a consistent interaction phenomenon exceeding the threshold is found. The threshold is set by counting the extreme values ​​and calculating the interquartile range in 10,000 advertising records. When the data distribution Ads outside of Q1 and Q3 that exceed 1.5 times the Q3-Q1 range are considered extreme. Finally, click-through rates, comment rates, and dwell time are measured side-by-side across platforms. If some platforms show significantly higher clickthrough rates for specific copy or color schemes, while other platforms experience greater peak engagement during nighttime than during daytime hours, this indicates significant disparity in audience engagement across these platforms. Once all comparisons are complete, review the corresponding demographic profiles and conduct a cross-analysis of response levels across the same age groups or interest tags. Map each platform's performance results to the original ad effectiveness evaluation results, integrating the weighted ratios of clicks and comments to rank the platforms. Once the rankings are determined, further comparisons can be made between user behavior characteristics of high- and low-ranking platforms. These differences can be recorded and summarized to reveal performance differences across platforms. Combined with feedback from different audience groups on different platforms, platforms can be prioritized based on the previously determined click and feedback ranges. Specific adaptation instructions can be made for ad delivery durations and creative styles to generate an optimized ad delivery strategy.

[0127] Based on the advertising delivery strategy optimization plan, we first adjust the advertising delivery time of each platform and re-divide the display segments, defining the period from 6:00 to 9:00 in the morning as Segment A, 9:00 to 18:00 in the daytime as Segment B, and 18:00 to 24:00 in the evening as Segment C. Based on the previous statistics of the active peaks of each platform, we determine which segments are most suitable for delivering flagship product ads and which segments are more suitable for displaying emotional appeal content. After confirming each segment, we associate the copy of the advertising content with the visual material. For example, we allocate long narrative videos to social platforms that prefer videos, and focus on key headlines and a small number of pictures on information platforms that prefer short text reading. When setting the budget allocation ratio, we first search for the difference between the previous delivery cost and the interactive income. If the difference is between 50% and 80%, it means that the investment can continue to increase. This range is based on the statistical median of advertising expenditure and recovery income. When the actual delivery rate exceeds 80% of the median, it is recommended to lower the delivery budget to control marginal investment. If the difference is less than 20%, it means that the investment is relatively conservative and the budget can be appropriately increased. After the budget allocation and display methods of all platforms have been adjusted accordingly, the color scheme of each advertisement in the corresponding time period must also be cross-examined to confirm that the color contrast meets the previously collected readability standards. Before the advertisement goes online, the display frequency and delivery content of each platform are rearranged, and the number of words in the ad title is limited, for example, within 15 to 30 words. All the above configurations are written into the centralized data, and finally the delivery schedule and the corresponding method of advertising materials are merged to form an optimized advertising deployment plan.

[0128] The steps to obtain the delivery results are:

[0129] According to the optimized advertising deployment plan, advertisements are placed, user behavior data after advertisement exposure is monitored, and real-time advertisement feedback data is generated;

[0130] Based on the real-time advertising feedback data, the advertising delivery results are calculated using the expression:

[0131]

[0132] in, Representative Advertising The advertising results, Representative Advertising In time User page interaction intensity within Representative Advertising Depth of content browsing, Representative Advertising Page bounce rate, Representative Advertising The average length of stay in each time period, Representative Advertising The number of repeated visits by users, T represents the total duration after the ad is delivered;

[0133] Based on the advertising delivery results, analyze the user response to the advertisement on each platform and time period to obtain the delivery results.

[0134] Specifically, according to the optimized advertising deployment plan obtained above, the exposure position and display period of each advertisement are sorted out at the beginning of delivery, and the delivery time of the same advertisement on different platforms is matched with the crowd information. At the same time, the page opening status and interactive click mark under each exposure are recorded. If the exposure volume within a certain delivery period is greater than 100 times and the number of clicks is less than 3 times, it is marked as a low conversion period. The values ​​of 100 and 3 are derived from the lower quartile obtained by statistics of all delivery records in the past seven days. For periods with obvious disconnection between clicks and jumps, special attention should be paid to the user browsing duration. When the duration is less than 5 seconds, it will be included in the list of quickly jumped records and the advertising type and audience characteristics will be compared again. At the same time, attention should be paid to collecting the user's browsing time in each The scrolling operation track and material playback progress after opening the advertising page for the first time are collected, and all browsing actions are associated with the start time of the delivery. If there are repeat visitors, they will be identified and merged in the background. During the integration, it is also necessary to check whether there are continuous visits or stays exceeding 60 seconds. The 60-second threshold is set after summarizing the average playback completion rate of video ads. If it is found that a user has visited the advertising page more than 3 times and the visit duration is higher than 30 seconds each time, it will be recorded as a high-attention item. After the delivery is completed, the browsing and interaction information of all users on that day will be summarized by time period, and combined with the actual number of clicks and jump depth to form the original data sequence. Finally, the exposure, click status and browsing depth of the time period are confirmed to generate real-time advertising feedback data.

[0135] The benefit of this formula lies in combining the integral value of user interaction intensity over time with the difference between content browsing depth and page bounce rate, and integrating the average dwell time and number of repeat visits into the denominator, allowing multiple key behavioral factors to be balanced within a unified calculation framework.

[0136] The steps to obtain the parameter are as follows: In time To determine the user page interaction intensity within a certain period, it is necessary to quantify the click actions, page scrolling amplitude, video playback progress, and number of slides generated by users while browsing the ad page and accumulate them in each time segment. If a day is divided into 24 time periods, the interaction actions in each time period are counted, and then the counts of all time periods are combined to form an interaction intensity function with time as the horizontal axis. Then, during the monitoring period, the function is integrated over time to obtain a total intensity value. For example, during a 24-hour monitoring period of an advertisement, a total of 400 click behaviors, 320 scroll behaviors, and 120 video playback starts are recorded. A weighted mapping method is used to set the proportion of clicks, scrolls, and playbacks. Clicks are mapped to a single contribution value of 1.2, scrolls are mapped to 1.0, and playbacks are mapped to 0.8. After integration, the interaction value of the period is obtained. This is repeated through all periods to generate , which can be finally obtained in the integral operation For example, the total value of each time period in a day is 600.

[0137] The steps to obtain the parameter are as follows, which represents the advertisement The content browsing depth is determined by placing several monitoring points on the advertising page to determine whether the user has browsed to the middle or bottom of the page. Each monitoring point is triggered once it is reached by scrolling, and the value is converted based on the user's stay time at the text or image position. For example, the page has three monitoring points, located at 20%, 50%, and 80% of the page height respectively. When the user triggers each monitoring point in turn, the corresponding depth increment is recorded, and the accumulation finally forms a depth value. If it is found that the user continues to stay for more than 5 seconds after reaching 80%, an additional depth coefficient of 0.1 is added. The entire measurement process records each visit, and the average browsing depth is obtained after the cumulative visits are completed. For example, statistics on browsing data of 1,000 users show that the average overall depth is 3.2, that is, .

[0138] The steps to obtain the parameter are as follows, which represents the advertisement The page bounce rate is first collected. The user's actions of closing or switching pages within a short period of time after clicking on the ad page are collected. If the stay time is less than 3 seconds, it is considered a quick bounce. The 3-second benchmark value comes from the threshold obtained by classifying 50,000 access records. The number of quick bounces is divided by the total number of visits to get the bounce rate, and an additional mark is added for cases with obvious quick return behavior. For example, the total number of visits in a certain advertising statistical period is 3,000, of which the number of quick bounces reaches 450, then the bounce rate = 450 / 3,000 = 0.15. If it is monitored that some users repeatedly enter and exit quickly, only one bounce is recorded, and the bounce rate is finally used as The value of the ad, such as =0.15.

[0139] The steps to obtain the parameter are as follows, which represents the advertisement For the average duration of stay in each time period, we first count the number of seconds users stay on the ad page one by one. At the end of the visit, we record the departure time minus the entry time to get the stay interval. We divide the stay time of all users by time period to get the average value. Finally, we weight the average values ​​of different time periods to form the overall average value of the entire delivery cycle. For example, the average stay time in the morning, noon and night time periods is calculated as 8 seconds, 10 seconds and 12 seconds respectively, and weighted by the number of visitors in each time period as 0.3, 0.4 and 0.3 to get the overall average value. In an actual measurement, the value was approximately 10.2 seconds.

[0140] The steps to obtain the parameter are as follows, which represents the advertisement To calculate the number of repeated visits by users, it is necessary to perform user-level deduplication statistics on all visits to the ad page. If it is detected that the same user ID visits the same ad page multiple times, the number of repeated visits to the ad by the user will be accumulated one by one. For example, if a user visits the ad page 5 times in two days, the number of repeated visits to the ad by the user will be recorded as 5. After adding up the number of such visits by all users, the number of smoothing is performed based on the total number of visitors. If the ad page has a total of 2000 visits in one cycle, of which 500 are repeated visits, then , the ratio or an integer form can be regarded as The final value of .

[0141] The parameter is obtained as follows: it represents the total duration of the ad after it is launched, that is, the time is counted from the time the ad is launched online, and it is pushed back in hours or minutes until the end of the launch or the end of this statistics, and then the duration is recorded as For example, if an ad starts running at 9:00 am and ends at 9:00 am the next day, for a total of 24 hours, then Hour.

[0142] Calculation process:

[0143] make , , , , , first calculate , add the integral value to the difference , then calculate = = = , and compare it with Add , and finally:

[0144]

[0145] The results show that the advertising The result of the delivery is about 57.45. When comparing values ​​in the same list, higher values ​​indicate more significant performance in dimensions such as user interaction intensity, browsing depth, and repeated visits. A value that is too low indicates that the data indicators of certain links are not ideal. At the same time, this value can be combined with other interaction dimensions obtained previously for cross-analysis. If it is greater than 50, it means that the overall activity of the ad is high during this period, and if it is less than 10, it means that the ad cannot obtain sufficient user interaction or repeat visits within the available time.

[0146] After calculating the advertising results, the The values ​​are listed by time period, and the visitor information of the advertisement in each time period is read and compared with the number of comments and page bounce records previously determined. If a high The corresponding visits of the value are also relatively large. In this case, a more in-depth browsing analysis is carried out during the period and the user's scrolling trajectory is checked to see whether it has reached the monitoring point mentioned above. At the same time, attention is paid to the stay pattern of repeat users and whether they are mainly distributed in specific time periods or platforms. All abnormal data information is compared with the threshold range and it is determined whether any period with a bounce rate significantly greater than 0.5 is related to content typesetting or excessive materials. When these time periods match the page layout, they are made into a report comparison. After confirming that all records have been associated, a time-based delivery comparison table is generated to illustrate the browsing characteristics and focus of each time period. Finally, this information is compared horizontally with other advertising delivery situations to obtain the delivery results.

[0147] The present invention provides an advertising device, comprising: a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory, so that the advertising device executes the above-mentioned artificial intelligence-based advertising method for device sales.

[0148] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An artificial intelligence-based advertising method for device sales, characterized in that: The following steps are involved: Collect social media behavior data and purchase records of consumers in the target market, and obtain standardized consumer data through data cleaning and formatting; Based on the standardized consumer data, conduct user profile analysis, including age, preferences, and purchasing power, to generate consumer segmentation results; Based on the consumer segmentation results, sentiment analysis is performed on the consumer groups to identify emotional reactions and preferences to the advertising content, thereby obtaining sentiment preference results; Adjusting advertising copy and visual elements based on the emotional preference results to generate customized advertising content; Using the customized advertising content, testing is performed on the platform, user interaction data and feedback are recorded, and advertising effectiveness evaluation results are obtained; based on the advertising effectiveness evaluation results, performance differences on different platforms are analyzed, advertising delivery strategies are optimized, and optimized advertising deployment plans are generated; Publish advertisements according to the optimized advertisement deployment plan, monitor real-time data streams, adjust advertisement content and delivery time according to the real-time data streams, and obtain delivery results; The steps for obtaining the emotion preference result are: Extracting advertising interaction data for each group from the consumer segmentation results, including the number of likes, number of comments, sharing frequency, and ad viewing time, and performing data screening and classification to obtain consumer group interaction characteristics; Based on the interaction characteristics of the consumer groups, the emotional response score of each group to the advertising content is calculated using the following formula: in, Representing a group The emotional response score, Representative groups Number of likes, Representative groups Number of comments, Representative groups Frequency of ad sharing, Representative groups ad viewing time; Based on the emotional response scores and in combination with the advertising interaction patterns of the consumer groups, the emotional trends of each consumer group towards the advertising content are analyzed to obtain emotional preference results; The steps for obtaining the customized advertising content are: Based on the emotional preference results, extract the preferences of different consumer groups for advertising text and visual design, and generate advertising copy features and visual element features; Based on the characteristics of the advertising copy and visual elements, screen the advertising text content and visual materials that meet the emotional preferences of the consumer group, adjust the expression of text sentences, the presentation of advertising themes and the structure of information transmission, and generate advertising copy adjustment plans and visual element adjustment plans; Based on the advertising copy adjustment plan and the visual element adjustment plan, the optimized copy content and visual materials are integrated, the matching degree between the advertising content and the characteristics of the consumer group is adjusted, and customized advertising content is generated; The steps for obtaining the advertising effect evaluation result are as follows: Using the customized advertising content, conduct advertising delivery tests on the platform, set advertising delivery strategies, including release time, display frequency, and target audience, collect user interaction behavior data, and generate a user interaction data set; Based on the user interaction dataset, the advertisement suitability score is calculated using the following formula: in, Indicates advertising The fitness score of Representative Advertising Number of clicks, Representative Advertising The residence time, Representative Advertising Number of shares, Representative Advertising Number of comments, Representative Advertising Number of impressions; Based on the advertisement suitability score, the performance of the advertisement on the platform is analyzed, and the user interaction effects of each advertisement are compared to obtain an advertisement effect evaluation result.

2. The device sales advertising method based on artificial intelligence according to claim 1, characterized in that: The steps for obtaining the standardized consumer data are: Collect social media behavior data and purchase records of consumers in the target market, remove duplicate data, correct data format errors, and obtain a preliminarily processed data set; Based on the preliminarily processed data set, missing values, outliers and data deviations are identified and processed, and data interpolation and outlier correction are applied to obtain standardized consumer data.

3. The device sales advertising method based on artificial intelligence according to claim 1, characterized in that: The steps for obtaining the consumer segmentation results are: Extracting the consumer's age, purchase frequency, and preferred category from the standardized consumer data to obtain a feature extraction result; Calculating the purchasing power index of each consumer based on the feature extraction results; Based on the purchasing power index, combined with consumers' purchase frequency, preferred categories and age range, consumers are categorized and a consumer segmentation result is generated.

4. The device sales advertising method based on artificial intelligence according to claim 1, characterized in that: The steps for obtaining the optimized advertisement deployment plan are as follows: Based on the advertising effectiveness evaluation results, calculate the performance differences of advertising on various platforms, analyze user behavior patterns on different platforms, compare the effectiveness of advertising content among different audience groups, and generate advertising strategy optimization plans; Based on the advertising delivery strategy optimization plan, adjust the advertising delivery time, adjust the display method of advertising content on various platforms, optimize the adaptation strategy of advertising materials, adjust the budget allocation ratio, and form an optimized advertising deployment plan.

5. The device sales advertising method based on artificial intelligence according to claim 1, characterized in that: The steps for obtaining the delivery results are: Delivering advertisements according to the optimized advertisement deployment plan, monitoring user behavior data after advertisement exposure, and generating real-time advertisement feedback data; Based on the real-time advertising feedback data, the advertising delivery result is calculated using the expression: in, Representative Advertising The advertising results, Representative Advertising In time User page interaction intensity within Representative Advertising Depth of content browsing, Representative Advertising Page bounce rate, Representative Advertising The average length of stay in each time period, Representative Advertising The number of repeated visits by users, T represents the total duration after the ad is delivered; Based on the advertisement delivery results, the user response to the advertisement on each platform and time period is analyzed to obtain the delivery results.

6. Advertising device, characterized in that, include: A processor and a memory, the memory being used to store a computer program, the processor being used to call and run the computer program stored in the memory, so that the advertising device executes the method according to any one of claims 1 to 5.

Images