Internet financial business information push method and system based on image recognition

Abstract

The present invention discloses an Internet financial business information push method and system based on image recognition, belonging to the technical field of financial data processing, step 1, financial business information classification: using an unsupervised algorithm to analyze and classify a collection of financial business advertisements, dividing the advertisements into different category groups, and obtaining a financial business advertisement category A; step 2, user classification: within a preset time range, using image recognition technology to collect user behavior data U relative to different financial business advertisements, using an unsupervised algorithm to analyze U, identifying user interest preferences and behavior patterns, dividing users into different category groups, and obtaining a number of user categories C; step 3, advertisement attractiveness evaluation: using an advertisement attractiveness evaluation model S to evaluate A, obtaining the average attractiveness of each advertisement category group to different user category groups; step 4, new user push. The invention improves the accuracy of information push.

Description

Technical Field

[0001] The present invention belongs to the technical field of financial data processing, and in particular to an Internet financial business information push method and system based on image recognition. Background Art

[0002] Internet financial business information push refers to the act of financial institutions or related service providers using Internet technology to send users information about financial products, service updates, market trends, risk warnings, etc. through email, SMS, mobile application push notifications, etc. This push service is designed to provide users with timely and personalized financial information to help users better understand and manage their financial status.

[0003] Existing information push usually classifies financial business information by attributes, for example, by product type. If the target user has viewed financial business information of a certain category, the same type of financial business information will continue to be pushed. Although this type of push has improved accuracy to a certain extent, it is still not accurate enough. This inaccuracy comes from the fact that people have diverse interests and are generally not interested in only one type of product. Therefore, mechanical type-based recommendations have limitations. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides an Internet financial business information push method and system based on image recognition. The present invention is implemented as follows: the Internet financial business information push method based on image recognition includes the following steps:

[0005] Step 1, classification of financial business information: using an unsupervised algorithm to analyze and classify the financial business advertisement set, classifying the advertisements into different category groups, and obtaining a financial business advertisement category A;

[0006] Step 2, user classification: within a preset time range, collect user behavior data U with respect to different financial service advertisements through image recognition technology, use unsupervised algorithms to analyze U, identify user interest preferences and behavior patterns, and classify users into different category groups to obtain several user categories C;

[0007] Step 3, advertisement attractiveness evaluation: Use advertisement attractiveness evaluation model S to evaluate A and obtain the average attractiveness of each advertisement category group to different user category groups;

[0008] Step 4, new user push: When a new user joins, collect his / her preliminary behavior data u_new, use the user classification model C to determine the category group c_new to which he / she belongs, and then recommend advertisements of the corresponding category group in the financial service advertisement set A to him / her based on c_new.

[0009] As a preferred embodiment of the present invention, in step 1, the classification of financial business information includes the following sub-steps:

[0010] Step 1.1, data cleaning of the financial business advertisement collection;

[0011] Step 1.2, using K-means clustering to analyze and classify the financial business advertisement collection;

[0012] Step 1.3, determining the characteristics of each financial business advertisement category group;

[0013] Step 1.4, obtain financial service advertisement category A, each category group corresponds to a preset category of user interests or needs.

[0014] As a preferred embodiment of the present invention, in step 1.2, using K-means clustering to analyze and classify the financial service advertisement set includes the following steps:

[0015] Step 1.2.1. Prepare the data set: collect a set of advertisements and check whether each advertisement has quantifiable features. If so, proceed to the next step. If not, collect a new set of advertisements until each advertisement has quantifiable features. The quantifiable features include at least one of the content description, style label, and target audience information.

[0016] Step 1.2.2. Feature extraction: extracting required features from the advertising data, including text analysis, image recognition, and audience analysis;

[0017] Step 1.2.3. Data preprocessing: Standardize or normalize the extracted features so that all features are on the same scale;

[0018] Step 1.2.4. Apply the K-means algorithm: set the K value, initialize the cluster center, and then iteratively assign each advertisement to the nearest cluster center and update the cluster center position until the cluster center no longer changes or the predetermined number of iterations is reached;

[0019] Step 1.2.5. Analyze clustering results: Analyze the clustering results obtained by the K-means algorithm and identify the type of advertisements represented by each cluster;

[0020] Step 1.2.6. Evaluate the model: Use the intra-cluster distance to evaluate the clustering effect, and adjust the K value or feature selection as needed to optimize the clustering results.

[0021] As preferred embodiment of the present invention, step 2 includes the following sub-steps:

[0022] Step 2.1 Set a time range, start image recognition technology, and monitor users’ interactions with financial business advertisements in real time;

[0023] Step 2.2 Collect user behavior data U, including clicks, browsing time, and conversion rate;

[0024] Step 2.3: Apply cluster analysis to process the collected behavioral data U;

[0025] Step 2.4 Analyze user behavior patterns and identify user interest preferences;

[0026] Step 2.5: According to user interest preferences, users are divided into different category groups C.

[0027] As a preferred embodiment of the present invention, step 3 includes: applying the advertisement attraction evaluation model S to evaluate each advertisement category group A, calculating its average attraction to different user category groups C, analyzing the evaluation results, and determining the attraction of each advertisement category group to the user;

[0028] Among them, the advertising attractiveness evaluation model S is based on user behavior data and advertising performance data.

[0029] As a preferred embodiment of the present invention, the construction of the advertisement attractiveness evaluation model S comprises the following steps:

[0030] Step 3.1. Obtain user behavior data, including at least one of the user's click rate, browsing time, and conversion rate;

[0031] Step 3.2. Obtain advertising performance data, including at least one of the number of ad exposures, number of clicks, and number of conversions;

[0032] Step 3.3. Evaluate user interest weights, and evaluate user interest weights for different advertising topics or product types based on user behavior data;

[0033] Step 3.4. Evaluate the weight of the advertisement's attractiveness to users based on the advertisement's performance data;

[0034] Step 3.5. Introduce a time decay factor to adjust the weights to describe the changes in user behavior and ad performance data over time;

[0035] The calculation formula of the advertisement attractiveness evaluation model S is as follows:

[0036] S = Σ (user interest weight i * ad attractiveness weight i * time decay factor i)

[0037] Here, i represents a combination of different user behavior data and advertising performance data.

[0038] As a preferred embodiment of the present invention, the time decay factor includes a time decay factor based on user behavior data and a time decay factor based on advertisement performance data. The time decay factor based on user behavior data is calculated as follows:

[0039] Set sub-time period t1 as the advertising time period, sub-time period t2 as the non-advertising time period before t1, and adjacent t2 and t1 as a unit time period;

[0040] Calculate the current cumulative total duration T0 of the preset category of users in the adjacent time periods t2 and t1;

[0041] Calculate the total accumulated average duration T1 of the current online preset category users;

[0042] Calculate the total accumulated average duration T2 of the preset category users crossing the dividing line within a continuous period t2 and t1;

[0043] The time decay factor based on user behavior data is one of T1 / T0, T2 / T0, and T2 / T1;

[0044] Or the time decay factor based on user behavior data is the weighted sum of T1 / T0, T2 / T0, and T2 / T1.

[0045] An Internet financial business information push system based on image recognition includes a central control module, and the central control module is signal-connected with:

[0046] Financial business information classification module: uses an unsupervised algorithm to analyze and classify the financial business advertisement collection, divides the advertisements into different category groups, and obtains the financial business advertisement category A;

[0047] User classification module: within a preset time range, collects user behavior data U with respect to different financial service advertisements through image recognition technology, analyzes U using an unsupervised algorithm, identifies user interest preferences and behavior patterns, and divides users into different category groups to obtain a number of user categories C;

[0048] Advertisement attractiveness evaluation module: Use the advertisement attractiveness evaluation model S to evaluate A and obtain the average attractiveness of each advertisement category group to different user category groups;

[0049] New user push module: When a new user joins, its preliminary behavior data u_new is collected, and the user classification model C is used to determine the category group c_new to which it belongs. Then, based on c_new, advertisements of the corresponding category group in the financial business advertisement set A are recommended to it.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] Improved the accuracy and efficiency of advertising push for financial services. Through the dual classification of advertisements and users by unsupervised algorithms, the system can more accurately understand the content of advertisements and the interests and preferences of users, thereby achieving accurate matching of advertisements with users. This can not only improve user satisfaction and acceptance of advertisements, but also significantly improve the accuracy of advertising information push, increase click-through rate and conversion rate, and bring higher market returns to financial institutions. In addition, the process is highly adaptable and scalable, and can continuously optimize and adjust the recommendation strategy as user behavior changes and advertising content is updated, ensuring that users are always provided with advertising content that best meets their needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flowchart of a method for pushing Internet financial business information based on image recognition provided by an embodiment of the present invention;

[0053] Figure 2 is a schematic diagram of an advertisement attractiveness evaluation model S provided by an embodiment of the present invention;

[0054] Figure 3 is a flowchart of financial business information classification provided by an embodiment of the present invention;

[0055] Figure 4 It is a flowchart of analyzing and classifying a set of financial service advertisements using K-means clustering provided by an embodiment of the present invention;

[0056] Figure 5 It is a flowchart of advertisement attractiveness evaluation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to further understand the content, features and effects of the present invention, the following embodiments are given as examples and described in detail with reference to the accompanying drawings.

[0058] The structure of the present invention is described in detail below in conjunction with the accompanying drawings.

[0059] like Figures 1 to 5 As shown, the Internet financial business information push method based on image recognition provided by the embodiment of the present invention includes the following steps:

[0060] Step 1, classification of financial business information: using an unsupervised algorithm to analyze and classify the financial business advertisement set, classifying the advertisements into different category groups, and obtaining a financial business advertisement category A;

[0061] Step 2, user classification: within a preset time range, collect user behavior data U with respect to different financial service advertisements through image recognition technology, use unsupervised algorithms to analyze U, identify user interest preferences and behavior patterns, and classify users into different category groups to obtain several user categories C;

[0062] Step 3, advertisement attractiveness evaluation: Use advertisement attractiveness evaluation model S to evaluate A and obtain the average attractiveness of each advertisement category group to different user category groups;

[0063] Step 4, new user push: When a new user joins, collect his / her preliminary behavior data u_new, use the user classification model C to determine the category group c_new to which he / she belongs, and then recommend advertisements of the corresponding category group in the financial service advertisement set A to him / her based on c_new.

[0064] Where U={u1, u2, ..., un}, each ui represents a user behavior data sample;

[0065] C={c1, c2, ..., cm}, each ci represents a user category;

[0066] A={a1, a2, ..., ak}, each ai represents a financial business advertisement sample;

[0067] Assume that there is a user behavior data set U, which contains 1000 user behavior data samples. Use an unsupervised algorithm to analyze these data and obtain a user classification model C, which divides users into 5 categories: C={c1,c2,c3,c4,c5}.

[0068] At the same time, there is a financial business advertisement collection A, which contains 200 financial business advertisement samples. These advertisements are analyzed and classified by unsupervised algorithms to obtain a financial business advertisement classification model, which divides the advertisements into 4 category groups: A={a1,a2, a3, a4}.

[0069] Next, we construct an advertising attractiveness evaluation model S. For example, we obtain the following partial evaluation results:

[0070]

[0071] Now, suppose a new user joins and its preliminary behavior data u_new is collected. Through the user classification model C, it is judged that this new user belongs to the user category group c3. Based on this judgment, the advertisements of the advertisement category group a3 in the financial service advertisement set A are recommended to the new user, because according to the evaluation model S, the advertisement category group a3 has the highest attractiveness score (8.5 points) to the user category group c3. In this way, financial service advertisements that are more in line with their interests and needs can be provided to new users, thereby improving the targeting of advertisements and user satisfaction.

[0072] Specifically, in step 1, the classification of financial business information includes the following sub-steps:

[0073] Step 1.1: Clean the data of the financial business advertisement collection to ensure the quality of analysis; for example, remove duplicate advertisement samples, correct erroneous information, ensure the accuracy and completeness of advertisement content, and provide a high-quality data foundation for subsequent analysis.

[0074] Step 1.2: Use K-means clustering to analyze and classify the financial business advertisement collection. Apply the K-means clustering algorithm to analyze the various features in the advertisement collection and classify the advertisements into different categories based on factors such as advertisement content, style, and target audience.

[0075] Step 1.3, determine the characteristics of each financial business advertising category group, such as advertising theme, product type, etc. For example, analyze the advertising samples in each clustering result and extract common features, such as the advertisements of a certain category group may mainly revolve around the theme of stock investment, while another category group may focus on fund investment or credit card promotion, etc.

[0076] Step 1.4, obtain financial service advertisement categories A, each category group corresponds to the user interests or needs of the preset category. For example, through the above steps, four financial service advertisement categories A={a1, a2, a3, a4} are finally obtained, where a1 corresponds to stock investment advertisements, a2 corresponds to fund product advertisements, a3 corresponds to credit card promotion advertisements, and a4 corresponds to insurance product advertisements. Each category group is designed to meet the interests or needs of a specific user group.

[0077] Specifically, in step 1.2, using K-means clustering to analyze and classify the financial service advertisement set includes the following steps:

[0078] Step 1.2.1. Prepare the data set: collect a set of advertisements and check whether each advertisement has quantifiable features. If so, proceed to the next step. If not, collect a new set of advertisements until each advertisement has quantifiable features. The quantifiable features include at least one of the content description, style label, and target audience information.

[0079] Step 1.2.2. Feature extraction: Extract the required features from the advertising data, including text analysis (such as keyword frequency), image recognition (such as visual style features), and audience analysis (such as age, gender, and interest preferences);

[0080] Step 1.2.3. Data preprocessing: Standardize or normalize the extracted features so that all features are on the same scale so that the algorithm can process them effectively.

[0081] Step 1.2.4. Apply the K-means algorithm: set the K value (i.e., the number of clusters), initialize the cluster centers, and then iteratively assign each ad to the nearest cluster center and update the cluster center position until the cluster center no longer changes or the predetermined number of iterations is reached;

[0082] Step 1.2.5. Analyze clustering results: Analyze the clustering results obtained by the K-means algorithm and identify the type of advertisements represented by each cluster, such as one cluster represents funds and another cluster represents stocks.

[0083] Step 1.2.6. Evaluate the model: Use the intra-cluster distance to evaluate the clustering effect, and adjust the K value or feature selection as needed to optimize the clustering results.

[0084] To analyze and classify a collection of financial business advertisements using K-means clustering, the following are the specific steps and examples:

[0085] Step 1.2.1. Prepare the dataset: Assume that 100 financial business advertisements are collected, each with a content description, style label, and target audience information. For example, the content description of advertisement A is "Investing in Stocks", the style label is "Professional", and the target audience information is "Middle-aged Investors".

[0086] Step 1.2.2. Feature extraction: Extract the following features from the advertisement data:

[0087] Text analysis: Calculate the frequency of keywords, such as "stocks", "funds", "investment", etc.

[0088] Image recognition: Use image recognition technology to extract the visual style features of the advertisement, such as color, layout, etc.

[0089] Audience analysis: Extract characteristics such as age, gender, interest preferences, etc. based on target audience information.

[0090] Step 1.2.3. Data preprocessing: Standardize or normalize the extracted features. For example, standardize the age range from 20-60 years old to between 0-1.

[0091] Step 1.2.4. Apply the K-means algorithm: select a K value of 5, initialize the cluster centers, and then iteratively assign each ad to the nearest cluster center and update the cluster center position until the cluster center no longer changes or the predetermined number of iterations is reached.

[0092] Step 1.2.5. Analyze clustering results: Assume that after clustering is completed, the following clustering results are obtained:

[0093] Cluster 1: represents fund advertisements, with the main keywords being “fund” and “long-term investment”.

[0094] Cluster 2: represents stock ads, with the main keywords being “stock” and “high risk high return”.

[0095] The other clusters represent advertisements for different financial products such as insurance, loans, and savings.

[0096] Step 1.2.6. Evaluate the model: Use the intra-cluster distance to evaluate the clustering effect. It is found that the intra-cluster distance between cluster 1 and cluster 2 is small, indicating that the ads in these two clusters are highly similar. Based on the evaluation results, it may be necessary to adjust the K value or feature selection to optimize the clustering results.

[0097] According to the clustering results, targeted marketing plans can be developed for different categories of advertisements. For example, for fund advertisements in cluster 1, a marketing strategy targeting long-term investment can be designed, while for stock advertisements in cluster 2, a marketing strategy emphasizing short-term high returns can be designed.

[0098] Specifically, step 2 includes the following sub-steps:

[0099] 2.1 Set a time range, activate image recognition technology, and monitor in real time the user's interactive behavior with financial service advertisements; for example, set the time range to user behavior data within the past month, activate image recognition technology, and monitor in real time the length of time users click on financial service advertisements, browse advertisement pages, and the purchase conversion rate of users completing the financial products recommended by the advertisements.

[0100] 2.2 Collect user behavior data U, including clicks, browsing time, and conversion rate; for example, collect the number of times users click on ads, the average time they browse the ad page each time, and the conversion rate of users purchasing financial products through the ad page.

[0101] 2.3 Apply cluster analysis to process the collected behavior data U; for example, use the K-means clustering algorithm to perform cluster analysis on the collected user behavior data and classify users with similar behavior characteristics into the same category.

[0102] 2.4 Analyze user behavior patterns and identify user interest preferences. For example, by analyzing clustering results, it is found that a certain category of users tend to click on a specific type of financial product advertisements and spend a longer time browsing these advertisements, thereby identifying that this type of user has a higher interest preference for this type of financial product.

[0103] 2.5 According to user interest preferences, users are divided into different category groups C. For example, according to user interest preferences for different financial product advertisements, users are divided into five category groups: C={c1, c2, c3, c4, c5}, where users in group c1 prefer stock financial product advertisements, users in group c2 prefer fund advertisements, and so on.

[0104] Furthermore, step 3 includes: applying an advertising attractiveness evaluation model S to evaluate each advertising category group A, calculating its average attractiveness to different user category groups C, analyzing the evaluation results, and determining the attractiveness of each advertising category group to users; wherein the advertising attractiveness evaluation model S is based on user behavior data and advertising performance data.

[0105] Specifically, the construction of the advertisement attractiveness evaluation model S includes the following steps:

[0106] Step 3.1. Obtain user behavior data, including at least one of the user's click rate, browsing time, and conversion rate;

[0107] Step 3.2. Obtain advertising performance data, including at least one of the number of ad exposures, number of clicks, and number of conversions;

[0108] Step 3.3. Evaluate user interest weights, and evaluate user interest weights for different advertising topics or product types based on user behavior data;

[0109] Step 3.4. Evaluate the weight of the advertisement's attractiveness to users based on the advertisement's performance data;

[0110] Step 3.5. Introduce a time decay factor to adjust the weights to describe the changes in user behavior and ad performance data over time;

[0111] The calculation formula of the advertisement attractiveness evaluation model S is as follows:

[0112] S = Σ (user interest weight i * ad attractiveness weight i * time decay factor i)

[0113] Among them, i represents a combination of different user behavior data and advertising performance data. Through this formula, the average attractiveness of each advertising category group to different user category groups can be calculated.

[0114] For example, assume the following data:

[0115] User behavior data:

[0116] User click rate: 0.8

[0117] Viewing time: 120 seconds

[0118] Conversion rate: 0.05

[0119] Ad performance data:

[0120] Ad exposure times: 1000

[0121] Ad clicks: 100

[0122] Ad conversions: 10

[0123] User interest weight:

[0124] Interest weight for ad topic A: 0.7

[0125] Interest weight for ad topic B: 0.3

[0126] Ad attractiveness weight:

[0127] The attractiveness weight of ad theme A: 0.6

[0128] Attractiveness weight of ad theme B: 0.4

[0129] Time decay factor:

[0130] Time decay factor based on user behavior data: 0.9

[0131] Time decay factor based on ad performance data: 0.8

[0132] Calculate the average attractiveness of each advertising theme to users:

[0133] For Ad Theme A:

[0134] S_A = (user click rate * interest weight in ad theme A * time decay factor based on user behavior data) * (ad clicks / ad exposures * attractiveness weight of ad theme A * time decay factor based on ad performance data) = (0.8 * 0.7 * 0.9) * (100 / 1000 * 0.6 * 0.8) = 0.2016

[0135] For Ad Theme B:

[0136] S_B = (user click rate * interest weight in ad theme B * time decay factor based on user behavior data) * (ad clicks / ad exposures * attractiveness weight of ad theme B * time decay factor based on ad performance data) = (0.8 * 0.3 * 0.9) * (10 / 1000 * 0.4 * 0.8) = 0.00576

[0137] Through calculation, we can get:

[0138] The average attractiveness of advertising theme A is 0.2016; the average attractiveness of advertising theme B is 0.00576. Therefore, the average attractiveness of advertising theme A to users is higher than that of advertising theme B. When recommending advertisements to users, advertisements related to advertising theme A will be recommended first.

[0139] It should be noted that in the above content about the advertising attractiveness evaluation model S, "i" represents a combination of different user behavior data and advertising performance data. Specifically, when constructing the advertising attractiveness evaluation model S, it is necessary to comprehensively consider a variety of user behavior data (such as click-through rate, browsing time, conversion rate, etc.) and advertising performance data (such as exposure times, click times, conversion times, etc.). There will be various combinations between these different types of data, and "i" is used to refer to these different specific combinations. For example, in the calculation process, there may be a combination of "user click-through rate and ad click times" as a case of "i", and there may also be other combinations such as "browsing time and ad exposure times". By calculating the corresponding products (user interest weight i * advertising attractiveness weight i * time decay factor i) for these different combinations (i.e. different "i") according to the formula, and finally summing up the products corresponding to all these different combinations (Σ), the value of the advertising attractiveness evaluation model S is obtained, so as to measure the average attractiveness of each advertising category group to different user category groups.

[0140] The three "i"s in the formula have the same meaning. In the formula S = Σ (user interest weight i * ad attractiveness weight i * time decay factor i) of this ad attractiveness evaluation model S, "i" is an index mark. It means that for each specific combination of user behavior data and ad performance data, there is a corresponding user interest weight, ad attractiveness weight and time decay factor. For example, when considering the data combination of "user click-through rate and ad click count" as "i = 1", then the user interest weight 1, ad attractiveness weight 1 and time decay factor 1 are specifically for this combination to determine its contribution value in the ad attractiveness evaluation. When "i = 2" may represent the combination of "user browsing time and ad exposure count", the corresponding user interest weight 2, ad attractiveness weight 2 and time decay factor 2 will be taken according to the characteristics of this new combination, and the contribution of this combination to the ad attractiveness will be calculated according to the formula. In this way, by calculating and summing all possible "i" (that is, all different data combinations) according to the formula, the ad attractiveness can be comprehensively evaluated.

[0141] It should be noted that the time decay factor can be set to linear decay or exponential decay. The calculation of the time decay factor includes: analyzing the changing trends of user behavior and advertising performance over time, such as whether the user click rate and conversion rate decrease over time; setting a decay function according to the data change trend, which can reflect how the data weight decreases over time. Common decay functions include linear decay and exponential decay. Calculate the decay factor for each time point based on the decay function. For example, if exponential decay is used, the decay factor is inversely proportional to the exponent of time. Of course, it can also be set to a constant jump type, such as setting a certain time period to α1 and another time period to α2, which can be set in advance.

[0142] Considering that the website cannot play advertisements all the time, otherwise the website traffic will be reduced. For example, if advertisements are played in sub-time period t1, the attractiveness of the advertisements in time period t1 to the users in time period t1 can be detected; however, the interval of the sub-time period t1 is t2 (at this time, the content that attracts users, such as videos, is played). Although the users in time period t2 can be captured, there are no advertisements in time period t2, so it is difficult to infer the attractiveness of the advertisements to the users in time period t2. Based on this scenario, the time decay factor based on user behavior data is calculated as follows:

[0143] Set sub-time period t1 as the advertising time period, sub-time period t2 as the non-advertising time period before t1, and adjacent t2 and t1 as a unit time period;

[0144] Calculate the current cumulative total duration T0 of the preset category of users in the adjacent time periods t2 and t1;

[0145] Calculate the total accumulated average duration T1 of the current online preset category users;

[0146] Calculate the total accumulated average duration T2 of the preset category users crossing the dividing line (the dividing line between adjacent t2 and t1) within an adjacent t2 and t1;

[0147] The time decay factor based on user behavior data is one of T1 / T0, T2 / T0, and T2 / T1;

[0148] Or the time decay factor based on user behavior data is the weighted sum of T1 / T0, T2 / T0, and T2 / T1.

[0149] For example, the adjacent times t2 and t1 are 600 seconds and 200 seconds respectively, that is, the total duration of the adjacent times t2 and t1 is 800 seconds, and the current time is 650 seconds, which is within t1.

[0150] User c11 entered at 200 seconds and exited at 500 seconds; he was online for 300 seconds but did not cross the boundary;

[0151] User c13 entered at 400 seconds and exited at 610 seconds; he was online for 210 seconds and crossed the dividing line for 10 seconds;

[0152] User c23 entered at 620 seconds and exited at 640 seconds; he was online for 20 seconds and crossed the dividing line for 20 seconds;

[0153] User c16 entered at 450 seconds and is still online; he was online for 200 seconds and crossed the demarcation line for 50 seconds;

[0154] Then the current cumulative total duration T0 of the c1 category user in the adjacent t2 and t1 time periods is the sum of 300, 210, and 200, which is 710 seconds;

[0155] Since users c11 and c13 have logged out, only user c16 is currently online among users in category c1. Therefore, the total accumulated average duration T1 of users in category c1 currently online is 200 seconds.

[0156] T2=(10+50) / 2=30 seconds.

[0157] Working principle: First, the system defines a specific time period structure, including the sub-time period t1 for playing advertisements and the non-advertising sub-time period t2 before it, which together constitute a unit time period. Next, the system calculates the current cumulative total duration T0 of users in the preset category in the adjacent time periods t2 and t1, which reflects the overall attention of users to these two time periods. Then, the system calculates the total accumulated average duration T1 of the currently online users of the preset category and the total accumulated average duration T2 of the users crossing the dividing line of t2 and t1. These two indicators represent the average activity of the current online users and the activity of the users crossing the dividing line, respectively. Finally, based on these calculation results, the system determines the time decay factor, which can be any one of T1 / T0, T2 / T0, T2 / T1, or the weighted sum of these ratios. This design allows the system to flexibly adjust the time decay factor according to the dynamic changes of user behavior data to reflect the differences in the importance of user behavior in different time periods. T1 / T0 represents the ratio of the total accumulated average duration of the current online users of the preset category to the current accumulated total duration of the users of this category in the adjacent time periods t2 and t1. This ratio reflects the proportion of current active users in the total time, and can be used to measure the relative relationship between current activity and historical cumulative activity. T2 / T0 represents the ratio of the total accumulated average duration of users in the preset category who cross the dividing line to the current accumulated total duration of users in this category in the adjacent t2 and t1 time periods. This ratio reflects the relative relationship between the user's activity across key time points and the total accumulated activity in a specific time period. T2 / T1 represents the ratio of the total accumulated average duration of users in the preset category who cross the dividing line to the total accumulated average duration of users in the preset category who are currently online. This ratio reflects the relative relationship between the user activity across key time points among the current active users and the overall current activity.

[0158] Beneficial effects: By introducing a time decay factor based on user behavior data, the system can more accurately capture the behavioral characteristics of users in different time periods, thereby providing more accurate data support for applications such as advertising and content recommendation. Specifically, the time decay factor can reflect the difference in user attention to advertising and non-advertising time periods, helping advertisers to more accurately choose the timing of advertising; at the same time, it can also reveal the user's behavior patterns when crossing the dividing line, providing additional reference information for the content recommendation algorithm. In addition, the dynamic adjustment mechanism of the time decay factor enables the system to adapt to changes in user behavior data and maintain a keen perception of user behavior characteristics, thereby further improving the accuracy and effectiveness of the application.

[0159] Furthermore, the method for pushing Internet financial business information based on image recognition also includes the following steps:

[0160] Step 5: Use the collected data on user responses to ads to further refine the user behavior analysis model and ad classification model. Deep learning networks are used to identify more subtle differences in user preferences and ad effectiveness.

[0161] Step 6: Based on the updated results of user behavior analysis and advertising classification models, further divide user categories and financial business advertising samples, and push advertisements that are more in line with users' detailed preferences to increase advertising appeal and conversion rates.

[0162] Step 7: Continue to monitor user responses to push ads and collect feedback data, and use this data to adjust the user classification model and ad classification model to achieve more accurate information push. This may involve iterative model training and parameter optimization.

[0163] Step 8: Repeat steps 6 and 7 to continuously optimize the user and advertisement classification model until the pushed financial business information is most accurately matched with user needs. This requires a continuous feedback loop to ensure the dynamic adaptability and long-term effectiveness of the information push system.

[0164] On the basis of the above, we further segment user categories and financial business advertisement samples. We can classify them initially based on their reactions to any one or two advertisements, and then recommend the most attractive advertisements to users in this category. We then observe the user's behavior towards advertisements in this category, and then further classify them until we push the most accurate financial business information.

[0165] An Internet financial business information push system based on image recognition includes a central control module, and the central control module is signal-connected with:

[0166] Financial business information classification module: uses an unsupervised algorithm to analyze and classify the financial business advertisement collection, divides the advertisements into different category groups, and obtains the financial business advertisement category A;

[0167] User classification module: within a preset time range, collects user behavior data U with respect to different financial service advertisements through image recognition technology, analyzes U using an unsupervised algorithm, identifies user interest preferences and behavior patterns, and divides users into different category groups to obtain a number of user categories C;

[0168] Advertisement attractiveness evaluation module: Use the advertisement attractiveness evaluation model S to evaluate A and obtain the average attractiveness of each advertisement category group to different user category groups;

[0169] New user push module: When a new user joins, its preliminary behavior data u_new is collected, and the user classification model C is used to determine the category group c_new to which it belongs. Then, based on c_new, advertisements of the corresponding category group in the financial business advertisement set A are recommended to it.

[0170] Working principle of the present invention:

[0171] The process first classifies financial business advertisements through an unsupervised algorithm to ensure that advertisements can be effectively grouped according to their characteristics. Subsequently, image recognition technology is used to capture user behavior data over a certain period of time, and these data are analyzed through an unsupervised algorithm to identify user interest preferences and behavior patterns, thereby dividing users into different categories. After that, a special advertising attractiveness evaluation model is used to evaluate the matching degree between advertising categories and user categories, and the attractiveness of each advertising category to different user categories is obtained. Finally, when a new user joins, the system will collect their preliminary behavior data, determine their category through the established user classification model, and recommend the advertisement that best suits their interests.

[0172] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0173] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The method for pushing Internet financial business information based on image recognition is characterized in that: The following steps are involved: Step 1, classification of financial business information: using an unsupervised algorithm to analyze and classify the financial business advertisement set, classifying the advertisements into different category groups, and obtaining a financial business advertisement category A; Step 2, user classification: within a preset time range, collect user behavior data U with respect to different financial service advertisements through image recognition technology, use unsupervised algorithms to analyze U, identify user interest preferences and behavior patterns, and classify users into different category groups to obtain several user categories C; Step 3, advertisement attractiveness evaluation: Use advertisement attractiveness evaluation model S to evaluate A and obtain the average attractiveness of each advertisement category group to different user category groups; Step 4, new user push: When a new user joins, collect their preliminary behavior data u_new, use the user classification model C to determine the category group c_new to which they belong, and then recommend advertisements of the corresponding category group in the financial business advertisement set A to them based on c_new; Step 3 includes: applying an advertisement attractiveness evaluation model S to evaluate each advertisement category group A, calculating its average attractiveness to different user category groups C, analyzing the evaluation results, and determining the attractiveness of each advertisement category group to the user; wherein the advertisement attractiveness evaluation model S is based on user behavior data and advertisement performance data; The construction of the advertisement attractiveness evaluation model S comprises the following steps: Step 3.

1. Obtain user behavior data, including at least one of the user's click rate, browsing time, and conversion rate; Step 3.

2. Obtain advertising performance data, including at least one of the number of ad exposures, number of clicks, and number of conversions; Step 3.

3. Evaluate user interest weights, and evaluate user interest weights for different advertising topics or product types based on user behavior data; Step 3.

4. Evaluate the weight of the advertisement's attractiveness to users based on the advertisement's performance data; Step 3.

5. Introduce a time decay factor to adjust the weights to describe the changes in user behavior and ad performance data over time; The calculation formula of the advertisement attractiveness evaluation model S is as follows: S = Σ (user interest weight i * ad attractiveness weight i * time decay factor i) Here, i represents a combination of different user behavior data and advertising performance data; The time decay factor includes a time decay factor based on user behavior data and a time decay factor based on advertisement performance data. The time decay factor based on user behavior data is calculated as follows: Set sub-time period t1 as the advertising time period, sub-time period t2 as the non-advertising time period before t1, and adjacent t2 and t1 as a unit time period; Calculate the current cumulative total duration T0 of the preset category of users in the adjacent time periods t2 and t1; Calculate the total accumulated average duration T1 of the current online preset category users; Calculate the total accumulated average duration T2 of the preset category users crossing the dividing line within a continuous period t2 and t1; The time decay factor based on user behavior data is one of T1 / T0, T2 / T0, and T2 / T1; Or the time decay factor based on user behavior data is the weighted sum of T1 / T0, T2 / T0, and T2 / T1; T1 / T0 represents the ratio of the total accumulated average duration of current online users of the preset category to the current accumulated total duration of users of this category in the adjacent t2 and t1 time periods, which is used to measure the relative relationship between the current activity and the historical accumulated activity; T2 / T0 represents the ratio of the total accumulated average duration of users of the preset category who cross the dividing line to the current accumulated total duration of users of this category in the adjacent t2 and t1 time periods, which reflects the relative relationship between the activity of users who cross key time points and the total accumulated activity in a specific time period; T2 / T1 represents the ratio of the total accumulated average duration of users of the preset category who cross the dividing line to the total accumulated average duration of current online users of the preset category, which reflects the relative relationship between the activity of users who cross key time points among the current active users and the overall current activity.

2. The method for pushing Internet financial business information based on image recognition according to claim 1, characterized in that: In step 1, the classification of financial business information includes the following sub-steps: Step 1.1, data cleaning of the financial business advertisement collection; Step 1.2, using K-means clustering to analyze and classify the financial business advertisement collection; Step 1.3, determining the characteristics of each financial business advertisement category group; Step 1.4, obtain financial service advertisement category A, each category group corresponds to a preset category of user interests or needs.

3. The method for pushing Internet financial business information based on image recognition as claimed in claim 2, characterized in that: In step 1.2, using K-means clustering to analyze and classify the financial business advertisement set includes the following steps: Step 1.2.

1. Prepare the data set: collect a set of advertisements and check whether each advertisement has quantifiable features. If so, proceed to the next step. If not, collect a new set of advertisements until each advertisement has quantifiable features. The quantifiable features include at least one of the content description, style label, and target audience information. Step 1.2.

2. Feature extraction: extracting required features from the advertising data, including text analysis, image recognition, and audience analysis; Step 1.2.

3. Data preprocessing: Standardize or normalize the extracted features so that all features are on the same scale; Step 1.2.

4. Apply the K-means algorithm: set the K value, initialize the cluster center, and then iteratively assign each advertisement to the nearest cluster center and update the cluster center position until the cluster center no longer changes or the predetermined number of iterations is reached; Step 1.2.

5. Analyze clustering results: Analyze the clustering results obtained by the K-means algorithm and identify the type of advertisements represented by each cluster; Step 1.2.

6. Evaluate the model: Use the intra-cluster distance to evaluate the clustering effect, and adjust the K value or feature selection as needed to optimize the clustering results.

4. The method for pushing Internet financial business information based on image recognition according to claim 1, characterized in that: Step 2 includes the following sub-steps: Step 2.1 Set a time range, start image recognition technology, and monitor users’ interactions with financial business advertisements in real time; Step 2.2 Collect user behavior data U, including clicks, browsing time, and conversion rate; Step 2.3: Apply cluster analysis to process the collected behavioral data U; Step 2.4 Analyze user behavior patterns and identify user interest preferences; Step 2.5: According to user interest preferences, users are divided into different category groups C.

5. An Internet financial business information push system based on image recognition, including a central control module, characterized in that: The central control module signal connections are: Financial business information classification module: uses an unsupervised algorithm to analyze and classify the financial business advertisement collection, divides the advertisements into different category groups, and obtains the financial business advertisement category A; User classification module: within a preset time range, collects user behavior data U with respect to different financial service advertisements through image recognition technology, analyzes U using an unsupervised algorithm, identifies user interest preferences and behavior patterns, and divides users into different category groups to obtain a number of user categories C; Advertisement attractiveness evaluation module: Use the advertisement attractiveness evaluation model S to evaluate A and obtain the average attractiveness of each advertisement category group to different user category groups; New user push module: When a new user joins, its preliminary behavior data u_new is collected, and the user classification model C is used to determine the category group c_new to which it belongs. Then, based on c_new, advertisements of the corresponding category group in the financial business advertisement set A are recommended to it.

Images