A method and server for recommending experts based on product data from short video platforms

Abstract

A method and server for recommending influencers based on product data from a short video platform, relating to the field of big data analysis, includes: obtaining influencer data on a preset ranking list of a target product, extracting crowd portrait data and fan age distribution data of the influencer data as an original data set; filtering the original data set to obtain product portrait data of the target product; extracting feature dimensions from the portrait data of the influencer to be matched to obtain influencer portrait data of the influencer to be matched; calculating the difference between the influencer portrait data and the product portrait data of the target product to obtain a recommendation index of the influencer to be matched for the target product; sorting and screening the influencers to be matched based on the recommendation index, and displaying an influencer recommendation list including a ranking sequence of multiple influencers to be matched. Implementation of this application can accurately match influencers to products, thus avoiding unsatisfactory promotion effects to a greater extent.

Description

Technical Field

[0001] The present application relates to the field of big data analysis, and in particular to a method and server for recommending experts based on product data from a short video platform. Background Art

[0002] With the rapid development of short video platforms, livestreaming has become a crucial sales channel for e-commerce. Data indicates that my country's livestream e-commerce market will exceed 4.9 trillion yuan in 2023, with short video platforms accounting for the majority of this market share. In this rapidly growing market, businesses and brands need to accurately identify expert livestreamers who are suitable for their products to achieve optimal marketing results and sales conversions.

[0003] Currently, product promotion on short video platforms primarily relies on manual operations to match influencers. Operations teams analyze influencers' historical sales data, follower profiles, content styles, and other metrics, combining experience and judgment to select suitable influencers for product promotion. Meanwhile, some e-commerce platforms have developed basic influencer recommendation systems, primarily based on superficial data such as influencer follower numbers and engagement rates.

[0004] However, with the rapid growth in the number of influencers, manual screening methods are unable to cope with the ever-expanding scale of information; and matching systems that rely solely on surface data find it difficult to deeply analyze the compatibility between influencers and products, which can easily lead to unsatisfactory promotion effects. Summary of the Invention

[0005] This application provides an expert recommendation method and server based on product data from a short video platform, which is used to accurately match experts with products and avoid unsatisfactory promotion effects to a greater extent.

[0006] In the first aspect, the present application provides an expert recommendation method based on short video platform product data, which is applied to a server, and the method includes: obtaining the data of influencers who bring goods to the target product on a preset ranking list, extracting the crowd portrait data and fan age distribution data of the influencer data as the original data set; filtering the original data set to obtain the product portrait data of the target product including the crowd distribution characteristics and fan age distribution characteristics; extracting the feature dimension of the portrait data to be matched, and obtaining the expert portrait data of the expert to be matched; calculating the difference between the expert portrait data and the product portrait data of the target product, and obtaining the recommendation index of the expert to be matched for the target product; sorting and screening the influencers to be matched based on the recommendation index, and displaying an expert recommendation list including a ranking sequence of multiple influencers to be matched.

[0007] In the above embodiment, the server achieves precise matching between products and influencers by acquiring the data of influencers promoting the target product and extracting their profile features. This method, combined with difference calculation, ranks influencer recommendations. This multi-dimensional data analysis, which considers both the influencer's user group characteristics and the product's target audience attributes, helps identify the most suitable influencers for product promotion, improving sales conversion rates and reducing the trial-and-error costs for merchants in selecting influencers.

[0008] In combination with some embodiments of the first aspect, in some embodiments, the step of filtering the original data set to obtain product portrait data of the target product including population distribution characteristics and fan age distribution characteristics specifically includes: respectively calculating the data averages of the population portrait data and fan age distribution data in the original data set to obtain data benchmark values ​​in multiple dimensions; calculating the standard deviation of the data benchmark value, and determining the target value range in multiple dimensions based on the data benchmark value and the standard deviation; removing abnormal data in the original data set that does not belong to the target value range to obtain a standard data set; performing weight calculation on the data of each dimension in the standard data set to obtain product portrait data of the target product including population distribution characteristics and fan age distribution characteristics.

[0009] In the above embodiment, the server establishes more accurate product portrait data by filtering and calculating weights on the original data set, uses data baseline values ​​and standard deviations to identify abnormal data, obtains a standard data set by removing outliers, and then performs weight calculations to make the product portrait more real and reliable.

[0010] In combination with some embodiments of the first aspect, in some embodiments, the step of extracting feature dimensions of the portrait data to be matched and obtaining the talent portrait data of the talent to be matched specifically includes: obtaining the portrait data to be matched, and determining multiple dominant dimension data in a preset pre-sequence among multiple dimensions of the portrait data to be matched; normalizing the multiple dominant dimension data and performing weight assignment to obtain a standard dimension array; calculating the dominant proportion of data on each dimension in the standard dimension array, and determining the talent portrait data of the talent to be matched based on the standard dimension array and the dominant proportion.

[0011] In the above embodiment, the server extracts the advantageous dimensions from the portrait data to be tested and performs normalization processing through the feature dimension extraction method, thereby establishing a complete expert portrait system. Through normalization processing and weight distribution, the core advantageous features of the expert can be accurately identified.

[0012] In combination with some embodiments of the first aspect, in some embodiments, the difference between the expert portrait data and the product portrait data of the target product is calculated to obtain the recommendation index of the target product by the expert to be matched, specifically including: extracting the product matching data corresponding to the expert portrait data in the product portrait data, performing difference calculation between the expert portrait data and the product matching data to obtain an initial difference array; determining the difference weight array according to the weight ratio of each data in the expert portrait data and the product matching data; performing weighted summation on the initial difference array according to the difference weight array to obtain reliability data; scaling the data density of the reliability data to generate a recommendation index.

[0013] In the above embodiment, the server uses a difference calculation method to perform a multi-dimensional comparison between the expert portrait and the product portrait to obtain an accurate recommendation index. Through data weight, difference calculation and density scaling of reliability data, a more scientific expert recommendation ranking is achieved.

[0014] In combination with some embodiments of the first aspect, in some embodiments, before obtaining the data of influencers who sell goods for the target product on a preset ranking list and extracting the population portrait data and fan age distribution data of the influencer data as the original data set, the method also includes: constructing a sliding time series containing multiple preset time windows, and counting the sales data, sales data and user evaluation data of the target product in each preset time window; performing time series analysis on the sales data, sales data and user evaluation data, and identifying the life cycle stage and market performance trend of the target product through growth rate, volatility and seasonal characteristics; determining the optimal time period for selling goods and the characteristics of the target audience group based on the life cycle stage and market performance trend, combined with the product category characteristics and price range; determining the preset benchmark dimensions and dimension weights of the target product based on the optimal time period for selling goods and the characteristics of the target audience group.

[0015] In the above embodiment, the server analyzes product performance by constructing a sliding time series, identifies product life cycles and market trends, and provides an optimization strategy in the time dimension for expert recommendations. By analyzing sales volume, sales revenue, and user evaluation data, the server determines the optimal time to bring products and the target audience, thereby improving the accuracy of recommendations.

[0016] In combination with some embodiments of the first aspect, in some embodiments, it is characterized in that after the step of sorting and screening the experts to be matched based on the recommendation index and displaying the expert recommendation list including the ranking sequence of multiple experts to be matched, the method also includes: obtaining the historical sales data of each matched expert in the expert recommendation list, and calculating the sales success rate of the matched experts in different product categories; generating a historical performance score of the matched expert based on the sales success rate and the time decay model of the historical sales data; weightedly integrating the historical performance score with the recommendation index, and re-sorting the experts in the expert recommendation list to obtain a recommended optimization list.

[0017] In the above embodiment, the server achieves dynamic optimization of recommendation results by analyzing the historical sales data and success rate of influencers and combining it with the time decay model. It improves the accuracy of recommendations by analyzing the historical performance of influencers in different product categories.

[0018] In combination with some embodiments of the first aspect, in some embodiments, based on the time decay model of the success rate of bringing goods and historical data of bringing goods, the steps of generating the historical performance score of the matched influencer specifically include: grouping the historical data of bringing goods into multiple dimensions according to product categories, price ranges and time dimensions, and calculating the success rate of bringing goods and sales conversion rate of each grouping dimension combination in different time periods; based on the time decay model, assigning dynamically decreasing weight coefficients to the success rate of bringing goods and sales conversion rate in different time periods to obtain a bringing goods decay curve; determining the bringing goods experience coefficient based on the matching influencer's frequency distribution and sales performance in each product category and price range; determining the historical performance score of the matched influencer based on the bringing goods decay curve, the bringing goods experience coefficient, and the recent activity and fan interaction indicators of the matched influencer.

[0019] In the above embodiment, the server comprehensively evaluates the historical performance of influencers through multi-dimensional grouping and time decay models, and establishes a complete historical performance scoring system based on factors such as the frequency of bringing goods, sales performance, and recent activity, thereby improving the reliability of recommendation results.

[0020] In a second aspect, an embodiment of the present application provides a server, comprising: one or more processors and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code comprising computer instructions, the one or more processors calling the computer instructions to enable the server to execute the method described in the first aspect and any possible implementation of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer program product comprising instructions, which, when executed on a server, enables the server to execute the method described in the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium comprising instructions. When the instructions are executed on a server, the server executes the method described in the first aspect and any possible implementation of the first aspect.

[0023] It is understandable that the server provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of the present application. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects of the corresponding methods and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. By adopting a secondary integration method based on product dimension data to obtain product influencer data, and filtering and extracting features through population portraits and fan age distribution characteristics, a complete product influencer matching system can be established, effectively solving the problem of low matching degree caused by relying solely on simple data indicators for influencer screening in existing technologies, thereby achieving more accurate influencer recommendations and improving the matching effect between products and influencers.

[0026] 2. Due to the adoption of a difference calculation method based on difference calculation and weight distribution, the reliability data is calculated by combining the initial difference array and the difference weight array, so it is possible to accurately measure the degree of adaptation between the influencer and the product, effectively solving the problem of the inability to quantitatively evaluate the matching degree between the influencer and the product in the existing technology, and thus realizing a more scientific calculation of the recommendation index, making the influencer recommendation results more objective and reliable, and improving the accuracy and practical value of the recommendation.

[0027] 3. Due to the use of a scoring method based on historical sales data analysis and a time decay model, combined with a weighted fusion of sales success rate and historical performance, it is possible to dynamically optimize the expert recommendation results, effectively solving the problem of static solidification of recommendation results in existing technologies and the inability to reflect the real-time performance of experts, thereby achieving more flexible and accurate expert recommendations and improving the practicality and reliability of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a method for recommending experts based on product data on a short video platform in an embodiment of the present application;

[0029] Figure 2 This is another flowchart of the expert recommendation method based on short video platform product data in an embodiment of the present application;

[0030] Figure 3 This is a schematic diagram of the physical device structure of the server in an embodiment of the present application. DETAILED DESCRIPTION

[0031] The terms used in the following examples of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application, the singular expressions "a", "an", "above", "the", and "this" are intended to include plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations of one or more of the listed items.

[0032] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of the technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of this application, unless otherwise specified, "plurality" means two or more.

[0033] To facilitate understanding, here are a few technical terms. "Influencers" are creators with a certain level of influence and a large fan base on short video platforms, who are able to effectively guide their followers to purchase products through their content creation and promotional capabilities. They demonstrate expertise or unique charm in diverse fields, such as fashion, beauty, digital products, and food, and play a key role in driving sales through e-commerce by showcasing, introducing, and recommending products.

[0034] The "eight demographic groups" refer to eight typical user groups identified by short video platforms based on multi-dimensional data including social attributes, consumption behaviors, and interests. These groups are: small-town youth, urban blue-collar workers, small-town middle-aged and elderly people, genz, urban seniors, senior middle-class professionals, sophisticated mothers, and emerging white-collar workers. For example, small-town youth typically live in urban areas and prioritize value for money, with a strong interest in fashion, trends, and entertainment. Urban blue-collar workers, primarily employed in urban areas for manual or skilled labor, prioritize practicality and quality, with a high demand for daily necessities, tools, and equipment. These demographic categories facilitate precise target audience analysis and provide key insights for matching products with influencers.

[0035] For ease of understanding, the application scenarios of the embodiments of the present application are introduced below.

[0036] A clothing brand launched a new down jacket on an e-commerce platform and needed to quickly identify suitable influencers to promote it. The operations team discovered that while some influencers had already successfully promoted the product, they faced difficulties in identifying new influencers. For example, a fashion blogger with 2 million followers, while his follower profile showed a predominantly urban, white-collar female demographic of 25-35, seemed well-suited, actually had suboptimal sales performance. The team discovered that selecting influencers based solely on their follower profile and follower count was highly biased and incapable of accurately predicting sales performance. They needed a method that could scientifically analyze the compatibility between products and influencers based on established success stories.

[0037] In related technologies, a simple matching method based on influencer follower count, engagement rate, and historical sales data can be used to achieve preliminary screening of products and influencers. The following describes a scenario using the related technology's influencer recommendation method based on short video platform product data.

[0038] An e-commerce operations team used a traditional influencer screening method, relying primarily on simple data provided by influencer agencies and personal experience. When searching for a suitable influencer for a skincare product, they first screened for influencers whose followers were aged 25-35 and whose female audience accounted for over 80%. They then made manual judgments based on the influencers' recent sales data. However, this method presented significant challenges: different operators used inconsistent criteria, resulting in selected influencers meeting all required metrics but ultimately failing to achieve satisfactory sales results. For example, they found a beauty influencer whose follower profile perfectly met the requirements, but whose sales conversion rate was far below expectations, resulting in wasted marketing resources.

[0039] The expert recommendation method based on short video platform product data in the embodiment of this application achieves more accurate product expert matching through secondary integration of high-quality expert data and outlier filtering, which not only improves the success rate of product promotion but also reduces operating costs. The following describes the scenario using the expert recommendation method based on short video platform product data in this application.

[0040] A beauty brand used this solution to match influencers for its toner. First, the system collected data from the product's top ten best-selling influencers. Through algorithmic analysis, it determined that the product's core audience is sophisticated mothers and emerging professionals aged 24-35. The system automatically filtered out outliers, such as an influencer whose user profile deviated significantly from the mainstream despite significant sales volume. The system then matched this standardized product profile with potential influencers across multiple dimensions, taking into account factors such as demographic distribution and age characteristics, ultimately recommending a list of influencers with the highest matching scores. After actual collaboration, these influencers' sales conversion rates were generally over 30% higher than those selected using traditional methods.

[0041] It can be seen that the expert recommendation method based on short video platform product data in the embodiment of the present application can not only achieve accurate expert matching, but also effectively solve the instability problem caused by subjective judgment in traditional methods, thereby achieving an improvement in the sales effect.

[0042] For ease of understanding, the following describes the process of the method provided by this implementation in combination with the above scenario. Figure 1 , which is a flow chart of the expert recommendation method based on short video platform product data in an embodiment of the present application.

[0043] S101. Obtain the data of influencers who sell the target product on a preset ranking list, and extract the population portrait data and fans age distribution data of the influencers as the original data set.

[0044] Among them, the target product refers to the specific product that needs to be matched with an influencer; the preset ranking list refers to a data list that sorts the influencers according to indicators such as sales volume and sales revenue; the influencer data refers to the relevant data of the influencer who successfully sells the product; the population portrait data refers to the multi-dimensional data reflecting the characteristics of the influencer's fan group, including the proportion data of eight major population types such as small town youth, urban blue-collar workers, and sophisticated mothers; the fan age distribution data represents the age composition of the influencer's fans, including the proportion data of multiple fan age ranges such as 18-23 years old and 24-30 years old; the original data set is used to represent the unprocessed initial data sample set.

[0045] After receiving a request for influencer matching, the server first needs to obtain historical sales data for the target product as a basis. Specifically, the server first accesses the product sales database to obtain the data of the top ten influencers in terms of sales volume. It then extracts the fan profiles of these influencers from the influencer database, including the proportion of eight demographic groups and the age distribution of their fans. This data is then integrated into a raw dataset containing multiple dimensional features.

[0046] In some embodiments, initial data acquisition and organization can be achieved through a variety of methods: optionally, the top 10 influencers can be selected based on the sales data of similar products over the past 30 days; the real-time fan profile data of these influencers can be extracted; the data can be organized and stored in a unified format; and timestamps can be added to facilitate subsequent analysis. Optionally, multiple time windows can be set to separately count influencer sales data; a comprehensive score can be generated based on indicators such as influencer interaction rate; the data of the 10 influencers with the highest comprehensive scores can be selected; and their profile data can be extracted to form a sample set. It is understood that other methods can also be used to achieve initial data acquisition and organization, such as through different ranking rules or data collection methods, which are not limited here.

[0047] For example, the server obtains the data of the top 10 sales figures:

[0048] 1. In the eight demographic dimensions:

[0049]

[0050] 2. In terms of fans’ age:

[0051]

[0052]

[0053] S102: Filter the original data set to obtain product portrait data of the target product, including demographic distribution characteristics and fan age distribution characteristics.

[0054] Among them, the data baseline value refers to the average level of data in each dimension; the standard deviation is used to indicate the degree of dispersion of the data; the target value interval represents the valid range of the data; abnormal data refers to data points that deviate significantly from the normal distribution; and the standard data set represents the set of valid data after screening.

[0055] After obtaining the raw data set, the server needs to perform data filtering to ensure data quality and reliability. Specifically, the server first calculates the mean and standard deviation of each dimension to determine a reasonable data range. It then removes outliers, such as data where the fan profile deviates significantly from the mainstream. The retained data is then standardized. Finally, weights are calculated to determine the final proportion of each dimension, forming a standardized product profile.

[0056] Continuing with the above example:

[0057] 1. In the eight demographic dimensions, the processing is as follows:

[0058] a. Calculate the average:

[0059]

[0060] b. Calculate the standard deviation:

[0061]

[0062] c. Calculate the filter interval:

[0063] The data calculation method for the fan age dimension is the same as the above eight population dimensions, so it will not be repeated here.

[0064] It should be noted that during the data filtering process, the standard deviation of the data baseline is calculated to measure the degree of dispersion of the data in each dimension. Taking the eight demographic dimensions as an example, assuming that the proportion of small town youth in the original data set is [0.21, 0.23, 0.25, 0.28, 0.19, 0.26, 0.24, 0.22, 0.27, 0.20], the average value is first calculated as 0.235. Then, the square of the difference between each data and the average value is calculated, such as (0.21-0.235)2, (0.23-0.235)2, etc., these squared differences are summed and divided by the number of data, and then the square root is taken to obtain the standard deviation. If the standard deviation is 0.03, combined with the average value of 0.235, the target value range can be determined to be [0.205, 0.265]. In this way, abnormal data that obviously deviates from the normal distribution, such as 0.15 or 0.30, can be effectively identified and removed, ensuring the reliability of the data and laying the foundation for the subsequent construction of accurate product portraits.

[0065] S103: Extract feature dimensions from the portrait data of the expert to be matched, and obtain the expert portrait data of the expert to be matched.

[0066] Among them, the experts to be matched refer to potential cooperative experts whose matching degree needs to be evaluated; the portrait data to be tested refers to the original data features of the experts to be matched; the feature dimension extraction is to determine their advantageous dimension data, which represents the most distinctive data dimensions of the experts.

[0067] After obtaining the product profile, the server needs to standardize the candidate influencer data for matching. Specifically, the server first obtains the original influencer profile data; then identifies the top three most representative features; then normalizes and weights these features; and finally generates the standardized influencer profile data.

[0068] In some embodiments, feature extraction for expert profiles can be achieved through various methods: optionally, selecting the top three features in each dimension; calculating the relative contributions of these features; applying a 4:3:3 weighting distribution; and generating a final weighted array. Optionally, calculating the contribution of each dimension to the expert profile; selecting the three dimensions with the highest contributions; performing normalization; and calculating relative weights. It is understood that other methods for extracting and normalizing expert features can also be used, such as through different feature selection algorithms, which are not limited here.

[0069] Continuing with the above example, the profile data of a certain expert to be tested is:

[0070] Eight major groups:

[0071]

[0072] Fan age:

[0073]

[0074] For the eight major groups, the processing steps are as follows:

[0075] 1. Among the eight demographic categories, the top three are: small town youth (24.95%), urban blue-collar workers (19.44%), and sophisticated mothers (13.4%).

[0076] 2. Redistribution of proportions:

[0077] Small town youth: 24.95% / (24.95%+19.44%+13.44%)=0.4314;

[0078] Urban blue-collar workers: 19.44% / (24.95%+19.44%+13.44%)=0.3361;

[0079] Exquisite mother: 13.44% / (24.95%+19.44%+13.44%)=0.2324.

[0080] 3. Weight distribution in proportion:

[0081] Small town youth: 0.4314*0.4=0.1726;

[0082] Urban blue-collar workers: 0.3361*0.3=0.1008;

[0083] Exquisite mother: 0.2324*0.3=0.0697.

[0084] 4. Normalize the data to ensure that the sum of the proportions is 1:

[0085] Small town youth: 0.1726 / (0.1726+0.1008+0.0697)=0.5031;

[0086] Urban blue-collar workers: 0.1008 / (0.1726+0.1008+0.0697)=0.2938;

[0087] Exquisite mother: 0.0697 / (0.1726+0.1008+0.0697)=0.2032.

[0088] The calculation method for the fan age dimension is the same as the above eight demographic dimensions, and the data obtained are:

[0089] Fans aged 31-40: 0.6783; fans aged 24-30: 0.2314; fans aged 41-50: 0.0827. The calculation process is not detailed here.

[0090] S104: Calculate the difference between the expert portrait data and the product portrait data of the target product to obtain the recommendation index of the target product by the expert to be matched.

[0091] Among them, the difference calculation represents the process of measuring the difference between two portrait data; the recommendation index represents the final matching score.

[0092] After the server normalizes the influencer and product profiles, it calculates their matching. Specifically, the server first extracts data from the same dimensions for comparison; then calculates the difference between each dimension; then weights the dimensions based on their importance; and finally, uses a mathematical model to convert the results into an intuitive recommendation metric.

[0093] Continuing with the above example, the processing process is:

[0094] 1. Calculate the difference in corresponding dimensions:

[0095] Eight demographic dimensions:

[0096] Small town youth: 0.2795-0.5031=-0.2236;

[0097] Urban blue-collar workers: 0.0645-0.2938=-0.2293;

[0098] Exquisite Mom: 0.0043-0.2032=-0.1987.

[0099] Fans age dimension:

[0100] Fan age 31-40: 0.4223-0.6783=-0.2560;

[0101] Fan age 24-39: 0.1956-0.2314=-0.0358;

[0102] Fan age 41-50: 0.2124-0.0827=0.1297.

[0103] 2. Calculation of the difference ratio (the difference divided by the data value in the influencer portrait data):

[0104] Eight demographic dimensions:

[0105] Small town youth: -0.2236 / 0.5031=-0.4444;

[0106] Urban blue-collar workers: -0.2293 / 0.2938=-0.7805;

[0107] Exquisite Mom: -0.1987 / 0.2032=-0.9779.

[0108] Fan age dimension: The calculation process is omitted, and the results are -0.3774, -0.1547, and 1.568.

[0109] 3. Use the weight proportions after normalization of the data in the previous step to sum up:

[0110] Eight major groups = -0.4444*0.5031+-0.7805*0.2938+-0.9779*0.2032=-0.6516;

[0111] Fan age = -0.3774*0.6873+-0.1547*0.2314+1.568*0.0827=-0.1655.

[0112] 4. Calculate the recommendation level:

[0113] Recommendation level = (((-0.6516+1) / 2)^0.5)+0.2*(((-0.1655+1) / 2)^0.5)=0.5465=54.65%.

[0114] S105 , sorting and screening the experts to be matched based on the recommendation index, and displaying an expert recommendation list including a ranking sequence of multiple experts to be matched.

[0115] Among them, sorting and filtering means sorting the experts according to the level of recommendation; ranking sequence refers to the list of experts arranged in descending order of recommendation; and the expert recommendation list represents the recommendation results finally presented to the user.

[0116] After calculating the recommendation scores of all potential matchmakers, the server needs to generate a recommendation list to facilitate decision-making. Specifically, the server first sorts all the matchmakers in descending order of recommendation score; then, it may set a recommendation score threshold for screening; and finally, it generates a ranked list containing the basic information of the matchmakers and their recommendation scores.

[0117] In some embodiments, recommendation lists can be generated in a variety of ways: optionally, setting a minimum recommendation threshold; screening qualified influencers; sorting by recommendation level; adding influencer basic information; and generating a recommendation list. Optionally, categorizing by recommendation level; selecting a fixed number of influencers per level; adjusting the order based on other factors; and generating a multi-level recommendation list. It is understood that other methods can also be used to generate recommendation lists, such as using different sorting rules or display methods, which are not limited here.

[0118] For example, a sample recommendation list is:

[0119] Expert A: 90.5% recommendation rate, 2 million followers, and sales of 500,000 in the past 30 days;

[0120] Expert B: 85.3% recommendation rate, 1.5 million followers, and sales of 450,000 in the past 30 days;

[0121] …

[0122] In the above example, a quantifiable product profile standard was established by systematically analyzing and processing the data of the top ten influencers by product sales. In practical applications, this method can quickly screen out potential influencers who are most suitable for a product, significantly improving influencer screening efficiency. The following supplements the scenarios of this example.

[0123] A comprehensive e-commerce platform has further upgraded this solution, incorporating product lifecycle and seasonal analysis. For example, for a sunscreen, the system not only analyzes influencer matching but also combines data characteristics from the product's peak sales season (summer) and off-season (winter) to recommend different influencers for different periods. In summer, the system prioritizes outdoor sports and travel influencers; in winter, it tends to recommend beauty and skincare influencers, adjusting the matching algorithm weights accordingly. This dynamic optimization ensures more stable product sales throughout the year, increasing sales by an average of over 40%.

[0124] After combining the above scenarios, the following is a more detailed description of the process of the method provided by this implementation. Figure 2 , which is another flow chart of the expert recommendation method based on short video platform product data in an embodiment of the present application.

[0125] S201. Obtain the data of influencers who sell the target product on a preset ranking list, and extract the population portrait data and fans age distribution data of the influencers as the original data set.

[0126] Referring to step S101, the server collects the original data set. For example, the server obtains the data of the top 10 sales figures, 1. In the eight demographic dimensions:

[0127]

[0128] 2. In terms of fans’ age:

[0129]

[0130] In some embodiments, the server first constructs a sliding time series containing multiple preset time windows, and counts the sales data, sales data and user evaluation data of the target product in each preset time window; performs time series analysis on the sales data, sales data and user evaluation data, and identifies the life cycle stage and market performance trend of the target product through growth rate, volatility and seasonal characteristics; determines the optimal time period for bringing goods to the target product and the characteristics of the target audience group based on the life cycle stage and market performance trend, combined with the product category characteristics and price range; determines the preset benchmark dimensions and dimension weights of the target product based on the optimal time period for bringing goods to the target product and the characteristics of the target audience group.

[0131] Among them, the preset time window represents a time analysis unit of fixed length; the sliding time series refers to a continuously moving time analysis interval; the growth rate represents the speed of change of the data; volatility refers to the amplitude of change of the data; seasonal characteristics are used to represent the law of periodic changes; the life cycle stage represents the development stage of the product in the market; the market performance trend refers to the development direction of product sales; the optimal time interval for bringing goods is used to represent the time period most suitable for promotion; the characteristics of the target audience group represent the attributes of the potential purchasing group; the preset benchmark dimension refers to the standard feature dimension for evaluating products.

[0132] After acquiring the target product, the server conducts a comprehensive data analysis to determine the optimal promotion strategy. Specifically, the server first establishes multiple fixed-length time windows, which slide sequentially at preset intervals. Within each window, the server then collects and organizes sales volume, sales revenue, and user evaluation data. It then analyzes this data's growth rate, fluctuations, and cyclical changes to identify the product's current market stage and development trends. Combined with the product's category characteristics and price positioning, the server determines the optimal promotion time and target user group. Finally, based on these analysis results, it sets key dimensions for evaluating the product and assigns corresponding weights.

[0133] S202. Calculate the average values ​​of the crowd portrait data and the fan age distribution data in the original data set respectively to obtain the data benchmark values ​​in multiple dimensions.

[0134] Among them, the data average value represents the arithmetic average result of the data in each dimension; the data benchmark value refers to the standard reference value used for subsequent analysis;.

[0135] After acquiring the raw data set, the server needs to establish a baseline standard for data analysis. Specifically, the server first categorizes the raw data set into eight demographic groups and age ranges; then sums the data for each dimension; then divides the sum by the sample size to obtain the average; and finally, uses these averages as the baseline values ​​for each dimension for subsequent data processing.

[0136] Continuing with the above example: 1. In the eight demographic dimensions, the processing is as follows:

[0137] Calculate the average (data base value):

[0138]

[0139] 2. The data calculation method for the fan age dimension is the same as the above eight population dimensions and will not be repeated here.

[0140] S203: Calculate the standard deviation of the data reference value, and determine the target value ranges in multiple dimensions based on the data reference value and the standard deviation.

[0141] Among them, the standard deviation represents the statistical quantity of the degree of data dispersion; the target value interval refers to the range limit of valid data, including the upper safety interval and the lower safety interval; the upper safety interval represents the result of adding the standard deviation to the data baseline value; the lower safety interval represents the result of subtracting the standard deviation from the data baseline value.

[0142] After obtaining the data baseline, the server needs to determine the appropriate distribution range for the data. Specifically, the server first calculates the sum of squared deviations between each dimension's data and the baseline; then finds the average of the sum of squared deviations; then takes the square root to obtain the standard deviation; and finally, adds or subtracts the standard deviation from the baseline to determine the target value range for each dimension.

[0143] Continuing with the above example: 1. In the eight demographic dimensions, the processing is as follows:

[0144] a. Calculate the standard deviation:

[0145]

[0146] b. Calculate the filter interval based on the data baseline value and standard deviation:

[0147]

[0148]

[0149] 2. The data calculation method for the fan age dimension is the same as the above eight population dimensions and will not be repeated here.

[0150] S204: Remove abnormal data that does not belong to the target value range in the original data set to obtain a standard data set.

[0151] Among them, abnormal data refers to data points that exceed the target value range; the standard data set refers to the filtered valid data set.

[0152] After determining the target value range, the server needs to clean the raw data. Specifically, the server first checks whether the data for each dimension is within the target range; then marks outliers that fall outside the range; then removes these outliers from the dataset; and finally obtains the cleaned standard dataset.

[0153] In some embodiments, anomalous data cleaning can be achieved through various methods: optionally, traversing and checking data across all dimensions; marking data that falls outside a certain range; batch-deleting anomalous data; recording the amount of cleaned data; and saving a standard dataset. Optionally, the Z-score of each data point can be calculated; a Z-score threshold can be set; data that exceeds the threshold can be removed; data distribution characteristics can be verified; and a final dataset can be generated. It is understood that other methods for anomalous data cleaning can also be used, such as using different anomaly detection algorithms, which are not limited here.

[0154] S205. Perform weight calculation on the data of each dimension in the standard data set to obtain product portrait data of the target product including demographic distribution characteristics and fan age distribution characteristics.

[0155] Among them, weight calculation represents the process of determining the importance of data in each dimension; population distribution characteristics refer to the proportion distribution of the eight major population groups; fan age distribution characteristics represent the proportion distribution of age groups; product portrait data is used to represent the standardized feature description of the product.

[0156] After receiving the standard dataset, the server needs to construct a standardized product profile. Specifically, the server first calculates the new mean of each dimension's data; then normalizes the data; then redistributes the weights of each dimension based on the 100% sum; and finally generates a complete product profile that includes demographic distribution and fan age distribution.

[0157] Continuing with the above example: 1. In the eight demographic dimensions, the processing is as follows:

[0158] After screening, the data benchmark value is calculated again based on the standard data set, and the re-proportion of the data benchmark value (the weight ratio of each data) is determined as follows:

[0159]

[0160] 2. The data calculation method for the fan age dimension is the same as the above eight population dimensions and will not be repeated here.

[0161] The above-mentioned set of proportions is the product portrait data, which represents the standardized distribution of products.

[0162] S206: Obtain the portrait data to be tested of the person to be matched, and determine a plurality of advantageous dimension data in a preset preceding sequence among the multiple dimensions of the portrait data to be tested.

[0163] Among them, the portrait data to be tested represents the original feature data of the candidate talent; the dominant dimension data refers to the main feature dimensions with higher values; and the preset prefix sequence represents the first few digits after sorting by numerical value.

[0164] After constructing the product profile, the server analyzes the characteristics of the prospective influencer. Specifically, the server first obtains the prospective influencer's complete profile data; then sorts all dimension data in descending order; selects the top three dimensions as dominant dimensions; and finally records the specific values ​​and rankings of these dominant dimensions.

[0165] Continuing with the above example, the profile data of a certain expert to be tested is:

[0166] 1. Eight population dimensions:

[0167]

[0168] 2. Fans’ age dimension:

[0169]

[0170] The data of the advantage dimensions are the top three data. Among the eight population dimensions, the top three are: small town youth 24.95%, urban blue-collar workers 19.44%, and sophisticated mothers 13.40%.

[0171] S207: Normalize the data of multiple dominant dimensions and distribute weights to obtain a standard dimension array.

[0172] Among them, normalization refers to the process of converting data to a uniform scale; weight allocation refers to assigning importance coefficients to different dimensions; and the standard dimension array is used to represent the processed standardized data set.

[0173] After determining the dominant dimension, the server needs to perform normalization. Specifically, the server first converts the dominant dimension data into standard values ​​in the [0, 1] interval; then assigns weights in a 4:3:3 ratio; then multiplies the standard values ​​by the weights; and finally, the server obtains a normalized dimension array.

[0174] Continuing with the above example, for the eight demographic dimensions:

[0175] 1. Redistribution (normalization):

[0176] Small town youth: 24.95% / (24.95%+19.44%+13.44%)=0.4314;

[0177] Urban blue-collar workers: 19.44% / (24.95%+19.44%+13.44%)=0.3361;

[0178] Exquisite mother: 13.44% / (24.95%+19.44%+13.44%)=0.2324.

[0179] 2. Weight distribution in proportion:

[0180] Small town youth: 0.4314*0.4=0.1726;

[0181] Urban blue-collar workers: 0.3361*0.3=0.1008;

[0182] Exquisite mother: 0.2324*0.3=0.0697.

[0183] The standard dimension array for the eight major groups is [0.1726, 0.1008, 0.0697]; a similar calculation is performed on the age of fans, and the result is [0.2450, 0.0863, 0.0299].

[0184] S208. Calculate the dominant ratio of data on each dimension in the standard dimension array, and determine the expert portrait data of the expert to be matched based on the standard dimension array and the dominant ratio.

[0185] Among them, the advantage ratio represents the weight ratio of each dimension in the whole; the expert portrait data refers to the standardized feature description of the expert.

[0186] After receiving the standard dimension array, the server needs to construct a standard profile of the influencer. Specifically, the server first calculates the proportion of each dimension's data in the total; then uses this proportion as the feature weight for that dimension; then, combined with the standard dimension array, generates a complete feature description; and finally, forms the standardized profile data of the influencer.

[0187] Continuing with the above example, for the eight demographic dimensions, to ensure that the sum of the proportions is 1, the advantage proportions are calculated as follows:

[0188] Small town youth: 0.1726 / (0.1726+0.1008+0.0697)=0.5031;

[0189] Urban blue-collar workers: 0.1008 / (0.1726+0.1008+0.0697)=0.2938;

[0190] Exquisite mother: 0.0697 / (0.1726+0.1008+0.0697)=0.2032;

[0191] That is, the weight ratios of the eight major groups are [0.5031, 0.2938, 0.2032]; the weight ratios and the standard dimension array [0.1726, 0.1008, 0.0697] together constitute the portrait of the expert in the eight major group dimensions.

[0192] Similarly, in the fan age dimension, the standard dimension array is [0.2450, 0.0863, 0.0299], and the weight ratio is [0.6783, 0.2314, 0.0827].

[0193] S209: Extract the product matching data corresponding to the expert portrait data from the product portrait data, calculate the difference between the expert portrait data and the product matching data, and obtain an initial difference array.

[0194] Among them, the product matching data represents the product data of the dimension corresponding to the influencer portrait; the initial difference array is used to represent the unweighted original difference value.

[0195] After obtaining the influencer profile data, the server needs to calculate its match with the product profile. Specifically, the server first extracts data from the product profile that has the same dimensions as the influencer profile; then, it compares the two sets of data one-to-one; then, it calculates the difference in each dimension; and finally, it forms an initial difference array that contains all the differences in the dimensions.

[0196] Continuing with the above example, calculate the difference in the corresponding dimensions:

[0197] a. Eight major demographic dimensions:

[0198] Small town youth: 0.2795-0.5031=-0.2236;

[0199] Urban blue-collar workers: 0.0645-0.2938=-0.2293;

[0200] Exquisite Mom: 0.0043-0.2032=-0.1987.

[0201] The initial difference array of the eight population dimensions is [-0.2236, -0.2293, -0.1987].

[0202] b. Fans’ age dimension:

[0203] Fan age 31-40: 0.4223-0.6783=-0.2560;

[0204] Fan age 24-39: 0.1956-0.2314=-0.0358;

[0205] Fan age 41-50: 0.2124-0.0827=0.1297.

[0206] The initial difference value array of the fan age dimension d1 is [-0.2560, -0.0358, 0.1297].

[0207] S210. Determine a difference weight array based on the weight proportions of each data in the expert portrait data and the product matching data.

[0208] Among them, the weight ratio indicates the importance of each dimension in the overall feature; the difference weight array refers to the weight coefficient used to adjust the importance of the difference.

[0209] After obtaining the initial difference, the server needs to consider the importance of each dimension. Specifically, the server first obtains the dimension weights of the influencer and product profiles, then calculates the combined weight of each dimension, normalizes the weights, and finally generates a weight array for difference calculation.

[0210] Continuing with the above example, the difference ratio calculation (the difference divided by the data value in the influencer portrait data) is:

[0211] Eight demographic dimensions:

[0212] Small town youth: -0.2236 / 0.5031=-0.4444;

[0213] Urban blue-collar workers: -0.2293 / 0.2938=-0.7805;

[0214] Exquisite mother: -0.1987 / 0.2032=-0.9779. Array [-0.4444, -0.7805, -0.9779].

[0215] Fan age dimension: The calculation process is omitted and the result is [-0.3774, -0.1547, 1.568].

[0216] S211 , performing weighted summation on the initial difference array according to the difference weight array to obtain reliability data.

[0217] The weighted summation refers to the accumulation of weighted differences; the reliability data refers to the final matching score; and the matching score is used to represent the numerical value of the overall matching degree.

[0218] After obtaining the difference weights, the server needs to calculate the final match degree. Specifically, the server first multiplies the difference by the corresponding weight; then sums all weighted differences; then normalizes them; and finally obtains the reliability data reflecting the match degree.

[0219] Continuing with the above example, we use the weighted proportions after normalizing the data in the previous step to sum up:

[0220] The reliability data of the eight population dimensions are as follows:

[0221] -0.4444*0.5031+-0.7805*0.2938+-0.9779*0.2032=-0.6516;

[0222] The fan age dimension is: -0.3774*0.6873+-0.1547*0.2314+1.568*0.0827=-0.1655.

[0223] S212: Scaling the data density of the reliability data to generate a recommendation index.

[0224] Among them, data density represents the distribution characteristics of reliability data; scaling refers to the process of interval conversion of data; recommendation index is used to represent the final matching score; numerical mapping refers to the process of converting data to a new interval; and score standardization refers to the process of unifying the score scale.

[0225] After receiving the reliability data, the server needs to convert it into a more intuitive recommendation index. Specifically, the server first performs a normal distribution test on the reliability data, then designs an appropriate conversion function, maps the data to a range of 0-100%, and finally generates a standardized recommendation index.

[0226] Continuing with the above example, the recommendation index of this expert is:

[0227] (((-0.6516+1) / 2)^0.5)+0.2*(((-0.1655+1) / 2)^0.5)=0.5465=54.65%.

[0228] S213: Sort and filter the experts to be matched based on the recommendation index, and display an expert recommendation list including a ranking sequence of multiple experts to be matched.

[0229] Referring to step S105, the server generates an expert recommendation list.

[0230] For example, a sample recommendation list is:

[0231] Expert A: 90.5% recommendation rate, 2 million followers, and sales of 500,000 in the past 30 days;

[0232] Expert B: 85.3% recommendation rate, 1.5 million followers, and sales of 450,000 in the past 30 days;

[0233] …

[0234] S214. Obtain the historical sales data of each matched expert in the expert recommendation list, and calculate the sales success rate of the matched experts in different product categories.

[0235] Among them, historical sales data represents the influencer’s past sales records; sales success rate refers to the completion ratio of product sales; product categories represent the classification of different types of products; sales performance is used to represent the influencer’s historical performance; and the data statistical period refers to the time range for collecting historical data.

[0236] After generating the initial recommendation list, the server needs to evaluate the influencer's historical performance. Specifically, the server first obtains the influencer's recent sales records; then categorizes and compiles statistics by product category; then calculates the success rate for each category; and finally generates a performance report for each category.

[0237] In some embodiments, the success rate of product sales can be calculated using a variety of methods: optionally, collecting sales data; classifying and analyzing sales data by category; calculating completion rates; weighting the data; and generating a success rate report. Alternatively, a statistical period can be set; valid data can be extracted; classified and summarized; conversion rates can be calculated; and evaluation results can be output. It is understood that other methods can also be used to calculate the success rate of product sales, such as using different statistical methods or evaluation criteria, which are not limited here.

[0238] S215. Generate a historical performance score for the matched influencer based on the sales success rate and the time decay model of historical sales data.

[0239] Among them, the time decay model refers to a calculation model that takes into account the influence of time factors; the historical performance score refers to the final score of comprehensive historical data; the time weight represents the importance of data in different periods; and the decay coefficient is used to indicate the degree of influence of time on data.

[0240] After determining the sales success rate, the server needs to consider the impact of time. Specifically, the server first sets a time decay function; then applies a decay coefficient to data from different periods; then calculates a weighted average score; and finally generates a comprehensive score reflecting historical performance.

[0241] It should be noted that the time decay model in this application uses historical sales data as a training set during training, including time series data such as product sales, sales revenue, conversion rate, etc., and optimizes the decay parameters by minimizing the prediction error. The training standard is to minimize the mean square error between the sales performance predicted by the model and the actual performance, while taking into account the importance differences of data in different time periods. The core of the time decay model is an exponential decay function, which contains two key parameters: the basic decay rate and the time interval. The model structure includes a time weight calculation layer, a data weighting layer, and an output normalization layer. The basic form is W(t) = α^(t / T), where α is the basic decay rate, t is the time interval, and T is the characteristic time scale. The input data of the time decay model include historical performance data such as the success rate of sales, sales conversion rate, and their corresponding timestamps; through model calculation, the weight coefficient of each time period can be obtained, and the sales decay curve can be output to characterize the timeliness contribution of the influencer's historical performance.

[0242] The core of the time decay model is the exponential decay function W(t) = α^(t / T). Assuming the basic decay rate α is 0.9 and the characteristic time scale T is 30 days, for the sales success rate data of a certain influencer 60 days ago, the time interval t is 60 days, and the weight coefficient W(60) = 0.9^(60 / 30)≈0.81 can be calculated. If the sales success rate in this period is 40% and the sales conversion rate is 30%, the weighted index value is 0.81*(0.4*0.5+0.3*0.5) = 0.2916 (assuming that the weights of the sales success rate and sales conversion rate are 0.5 each). At the same time, the sales experience coefficient is determined based on the sales frequency distribution and sales performance of the influencer in each commodity category and price range. For example, if a certain influencer has a high frequency of sales in the beauty category and has good sales performance, the sales experience coefficient is 1.2. Combined with recent activity (such as a comprehensive evaluation of indicators such as the number of videos released in the past 30 days and the length of live broadcasts) and fan interaction indicators (likes, comments, sharing rates, etc.), if the recent activity score is 0.8 and the fan interaction index score is 0.9, through weighted fusion (assuming that the weight of the experience coefficient for bringing goods is 0.4, the weight of recent activity is 0.3, and the weight of the fan interaction index is 0.3), the final historical performance score = 0.4*1.2+0.3*0.8+0.3*0.9=0.87, in order to comprehensively and dynamically evaluate the historical performance of the influencer and optimize the influencer recommendation results.

[0243] In some embodiments, the server will group historical sales data into multiple dimensions according to product categories, price ranges, and time dimensions, and calculate the sales success rate and sales conversion rate of each grouping dimension combination in different time periods; based on the time decay model, dynamically decreasing weight coefficients are assigned to the sales success rate and sales conversion rate in different time periods to obtain a sales decay curve; the sales experience coefficient is determined based on the sales frequency distribution and sales performance of the matched influencer in each product category and price range; based on the sales decay curve and sales experience coefficient, combined with the recent activity and fan interaction indicators of the matched influencer, the historical performance score of the matched influencer is determined.

[0244] Among them, multi-dimensional grouping means data classification according to multiple features; the success rate of sales refers to the proportion of successful sales; the sales conversion rate refers to the proportion of purchase intentions converted into actual sales; the time decay model is used to represent the calculation method of time impact; the sales decay curve represents the performance curve that changes over time; the sales frequency distribution refers to the frequency distribution of sales activities; the sales experience coefficient is used to indicate the professional level of the influencer; the recent activity indicates the frequency of the influencer's activities; the fan interaction index refers to the degree of interaction with fans.

[0245] After obtaining the influencer's historical data, the server conducts an in-depth performance analysis. Specifically, the server first categorizes and organizes the historical data by product type, price, and time period. It then calculates sales performance indicators for each combination over different time periods. A time decay model is then established, assigning decreasing weights to data from different time periods. The server then calculates a professional level coefficient based on the influencer's experience and sales performance. Finally, the server considers the influencer's recent performance and fan interaction to generate a final historical performance score.

[0246] S216. Perform weighted fusion of the historical performance score and the recommendation index, reorder the experts in the expert recommendation list, and obtain an optimized recommendation list.

[0247] Among them, weighted fusion refers to the calculation process that comprehensively considers two indicators; the recommended optimization list refers to the final expert ranking result; the fusion weight represents the importance of different indicators; the sorting rule is used to indicate the basis for expert ranking; and the optimization coefficient refers to the parameter used to adjust the final ranking.

[0248] After obtaining the historical performance scores and recommendation metrics, the server needs to generate the final recommendation ranking. Specifically, the server first sets the combined weights of the two metrics; then calculates the weighted composite score; then sorts all experts in descending order by score; and finally generates a recommendation optimization list containing complete information.

[0249] In the embodiment of the present application, due to the use of a secondary integration method based on product dimension data and a precise influencer matching algorithm, by systematically analyzing the user portraits of leading influencers in sales, processing outliers and calculating weights, it is possible to establish accurate product portrait standards and achieve precise matching with influencer data, effectively solving the problems of multiple subjective judgments, inconsistent standards and unstable effects in traditional influencer screening, thereby achieving an improvement in the effect of bringing goods and a significant improvement in operational efficiency, ensuring the accuracy and reliability of the recommendation results, enabling products to find the most suitable influencer resources more quickly, reducing marketing costs and improving the return on investment.

[0250] The following describes the server in the embodiment of the present invention from the perspective of hardware processing. Figure 3 , is a schematic diagram of a physical device structure of a server in an embodiment of the present application.

[0251] It should be noted that Figure 3 The structure of the server shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0252] like Figure 3 As shown, the server includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage part 308 to the random access memory (RAM) 303, such as executing the method described in the above embodiment. Various programs and data required for system operation are also stored in the RAM 303. The CPU 301, ROM 302 and RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0253] The following components are connected to the I / O interface 305: an input section 306 including an audio input device, a push button switch, and the like; an output section 307 including a liquid crystal display (LCD), an audio output device, an indicator light, and the like; a storage section 308 including a hard disk and the like; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. Removable media 311, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 310 as needed, so that computer programs read therefrom can be installed into the storage section 308 as needed.

[0254] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a computer-readable medium, the computer program including a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 309 and / or installed from a removable medium 311. When the computer program is executed by the central processing unit (CPU) 301, the various functions defined in the present invention are performed.

[0255] It should be noted that specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0256] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings.

[0257] Specifically, the server of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, the expert recommendation method based on the short video platform product data provided by the above embodiment is implemented.

[0258] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the server described in the above embodiments, or may exist independently and not be incorporated into the server. The storage medium carries one or more computer programs, which, when executed by a processor of the server, enable the server to implement the expert recommendation method based on short video platform product data provided in the above embodiments.

[0259] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0260] As used in the above embodiments, the term “when…” may be interpreted to mean “if…” or “after…” or “in response to determining…” or “in response to detecting…”, depending on the context. Similarly, the phrases “upon determining…” or “if (stated condition or event) is detected” may be interpreted to mean “if determining…” or “in response to determining…” or “upon detecting (stated condition or event)” or “in response to detecting (stated condition or event)”, depending on the context.

[0261] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for recommending experts based on commodity data on a short video platform, characterized in that: Applied to a server, the method includes: Obtain the data of influencers who are selling the target product on a preset ranking list, and extract the demographic data and age distribution data of the influencers as the original data set; Filtering the original data set to obtain product portrait data of the target product, including demographic distribution characteristics and fan age distribution characteristics; Extracting feature dimensions from the portrait data of the expert to be matched, and obtaining the expert portrait data of the expert to be matched; Calculate the difference between the expert portrait data and the product portrait data of the target product to obtain the recommendation index of the expert to be matched for the target product; Sorting and screening the experts to be matched based on the recommendation index, and displaying an expert recommendation list including a ranking sequence of multiple experts to be matched; Obtain the historical sales data of each matched expert in the expert recommendation list, and calculate the sales success rate of the matched experts in different product categories; Generate a historical performance score of the matched influencer based on the sales success rate and the time decay model of the historical sales data; The historical performance score is weightedly integrated with the recommendation index, and the experts in the expert recommendation list are reordered to obtain a recommendation optimization list.

2. The method according to claim 1, characterized in that The step of filtering the original data set to obtain product portrait data of the target product including demographic distribution characteristics and fan age distribution characteristics specifically includes: Calculate the average values ​​of the crowd portrait data and the fan age distribution data in the original data set respectively to obtain data benchmark values ​​in multiple dimensions; Calculating the standard deviation of the data reference value, and determining target value ranges in multiple dimensions based on the data reference value and the standard deviation; Removing abnormal data that does not belong to the target value range from the original data set to obtain a standard data set; The data of each dimension in the standard data set is weighted to obtain product portrait data of the target product including population distribution characteristics and fan age distribution characteristics.

3. The method according to claim 1, characterized in that The step of extracting feature dimensions from the test portrait data of the expert to be matched to obtain the expert portrait data of the expert to be matched specifically includes: Obtaining the portrait data of the person to be matched, and determining a plurality of dominant dimension data in a preset preceding sequence among the plurality of dimensions of the portrait data to be tested; Normalizing the plurality of dominant dimension data and performing weight distribution to obtain a standard dimension array; The dominant proportion of data on each dimension in the standard dimension array is calculated, and the expert portrait data of the expert to be matched is determined based on the standard dimension array and the dominant proportion.

4. The method according to claim 1, wherein The step of calculating the difference between the expert portrait data and the product portrait data of the target product to obtain the recommendation index of the to-be-matched expert for the target product specifically includes: Extracting product matching data corresponding to the expert portrait data from the product portrait data, performing difference calculation between the expert portrait data and the product matching data to obtain an initial difference array; Determine a difference weight array based on the weight proportions of each data in the expert portrait data and the product matching data; Performing weighted summation on the initial difference array according to the difference weight array to obtain reliability data; The data density of the reliability data is scaled to generate a recommendation index.

5. The method according to claim 1, wherein Before the step of obtaining the data of influencers who carry the target product on a preset ranking list and extracting the demographic data and the age distribution data of fans of the influencers as the original data set, the method further includes: Constructing a sliding time series containing multiple preset time windows, and collecting statistics on the sales volume data, sales revenue data, and user evaluation data of the target product in each of the preset time windows; Performing time series analysis on the sales volume data, the sales revenue data, and the user evaluation data to identify the life cycle stage and market performance trend of the target product through growth rate, volatility, and seasonality characteristics; Determine the optimal timeframe for selling the target product and the characteristics of the target audience based on the life cycle stage and market performance trends, combined with product category characteristics and price range; The preset benchmark dimensions and dimension weights of the target product are determined based on the optimal time period for bringing the product and the characteristics of the target audience group.

6. The method according to claim 1, characterized in that The step of generating the historical performance score of the matched influencer based on the product sales success rate and the time decay model of the historical product sales data specifically includes: The historical sales data is grouped into multiple dimensions according to product category, price range, and time dimension, and the sales success rate and sales conversion rate of each grouping dimension combination in different time periods are calculated; Based on the time decay model, dynamically decreasing weight coefficients are assigned to the sales success rate and the sales conversion rate in the different time periods to obtain a sales decay curve; Determine the sales experience coefficient based on the matched influencer's sales frequency distribution and sales performance in various product categories and price ranges; The historical performance score of the matched influencer is determined based on the product attenuation curve, the product experience coefficient, and the recent activity and fan interaction index of the matched influencer.

7. A server, characterized in that: The server includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the server to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on a server, the server is caused to execute the method according to any one of claims 1 to 6.

9. A computer program product, characterized in that When the computer program product is run on a server, the server is caused to perform the method according to any one of claims 1 to 6.

Images