A Big Data-Based Advertising Analysis Method and System

Abstract

The present invention discloses an advertising analysis method and system based on big data, including: when receiving an advertising push request for a specific user, first analyzing the interaction content of the user within the historical monitoring period to determine its interaction type. Then, using a pre-trained advertising interaction recommendation model, inferring the interaction type that best matches the user, and determining the advertising content to be pushed accordingly. Designed in this way, through in-depth mining of user behavior data, precise personalized advertising push is achieved, effectively improving the advertising effect and user experience.

Description

Technical Field

[0001] The present invention relates to the field of big data, and in particular, to an advertisement analysis method and system based on big data. Background Art

[0002] With the rapid development of Internet advertising, how to achieve accurate push has become a key challenge in the advertising industry. Traditional advertising push methods mostly rely on users' basic information and coarse-grained behavior data. Such methods often fail to fully capture users' personalized needs and interests, resulting in poor advertising push effects, poor user experience, and low advertising conversion rates. Summary of the Invention

[0003] The purpose of the present invention is to provide an advertisement analysis method and system based on big data.

[0004] In a first aspect, an embodiment of the present invention provides an advertisement analysis method based on big data, the method comprising:

[0005] When an advertisement push instruction for a target user account is obtained, obtaining the interaction content interacted by the target user account within a past monitoring period;

[0006] Determining a target interaction type matching the target user account according to the interaction content interacted by the target user account within a past monitoring period;

[0007] Loading the target interaction type into a pre-trained target advertisement interaction recommendation model, inferring a matching type of the target interaction type by the target advertisement interaction recommendation model, and determining the matching type of the target interaction type inferred and determined as the target inferred matching interaction type of the target interaction type;

[0008] Determining, from the interaction content corresponding to the target inferred matching interaction type, the interaction content to be pushed to the target user account.

[0009] In a second aspect, an embodiment of the present invention provides a server system, comprising a server, and the server is used to execute the method described in the first aspect.

[0010] Compared with the prior art, the beneficial effects provided by the present invention include: By adopting an advertisement analysis method and system based on big data disclosed in the present invention, when receiving an advertisement push request for a specific user, first analyze the interaction content of the user within a historical monitoring period to determine its interaction type. Then, use a pre-trained advertisement interaction recommendation model to infer the interaction type most matching the user, and determine the advertisement content to be pushed accordingly. Designed in this way, through in-depth mining of user behavior data, accurate personalized push of advertisements is achieved, effectively improving the advertisement effect and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative efforts.

[0012] Figure 1 It is a schematic flow chart of the steps of the advertising analysis method based on big data provided by an embodiment of the present invention;

[0013] Figure 2 It is a schematic block diagram of the structure of the computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0015] The following will describe in detail the specific implementation manners of the present invention in conjunction with the drawings.

[0016] To solve the technical problems in the foregoing background art, Figure 1 It is a schematic flow chart of the advertising analysis method based on big data provided by an embodiment of the present disclosure. The advertising analysis method based on big data will be introduced in detail below.

[0017] Step S201, when an advertisement push instruction for a target user account is obtained, obtain the interaction content interacted by the target user account within the past monitoring period;

[0018] Step S202, determine a target interaction type matching the target user account according to the interaction content interacted by the target user account within the past monitoring period;

[0019] Step S203, load the target interaction type into a pre-trained target advertisement interaction recommendation model, infer the matching type of the target interaction type by the target advertisement interaction recommendation model, and determine the matching type of the target interaction type inferred and determined as the target inferred matching interaction type of the target interaction type;

[0020] Step S204: Determine the interaction content to be pushed to the target user account from the interaction content corresponding to the target inference matching interaction type.

[0021] In an embodiment of the present invention, for example, when the server receives an advertisement push instruction for the user account "User123", it first retrieves all interaction records of "User123" in the past 30 days (set monitoring period) from the database. These interaction contents include but are not limited to the categories of products browsed, keywords searched, types of advertisements clicked, products purchased, etc. For example, the server finds that "User123" has browsed sports shoes and sportswear multiple times in the past 30 days and searched keywords such as "running shoes recommendation". Based on the collected interaction content, the server determines, through data analysis algorithms such as cluster analysis or association rule mining, that the main interest point of "User123" is in sports equipment, especially running shoes and sportswear. Therefore, the server determines "sports equipment" as the target interaction type. The server inputs the target interaction type of "sports equipment" into a pre-trained advertisement interaction recommendation model. This model can be a recommendation system based on deep learning, which predicts the content that users may be interested in by learning a large amount of historical interaction data of users. After receiving the input of "sports equipment", the model starts to infer the content that best matches this interaction type. The advertisement interaction recommendation model infers that the interaction types that best match "sports equipment" are "high-performance running shoes" and "breathable sportswear" by analyzing the historical data of "User123" and the behavior patterns of other similar users. These inference results are regarded as the target inference matching interaction types. Based on the results inferred by the model, the server selects advertisement contents related to "high-performance running shoes" and "breathable sportswear" from the advertisement library. For example, the server can select an advertisement for a newly launched high-performance running shoe and an advertisement for breathable quick-drying sportswear, and then push these two advertisements to "User123". Through the above steps, the server can accurately push the advertisement content that the user may be interested in to the target user account based on the big data-based advertisement analysis method.

[0022] In an embodiment of the present invention, the foregoing step S203 can be implemented through the following example.

[0023] Obtain the target type matching level for inferring the target inference matching interaction type;

[0024] Generate second model guidance information matching the target interaction type according to the target type matching level;

[0025] Load the second model guidance information into the target advertisement interaction recommendation model. Based on the second model guidance information, the target advertisement interaction recommendation model infers a matching type whose matching level with the target interaction type is the target type matching level, and determines the matching type whose matching level with the target interaction type is the target type matching level as the target inferred matching interaction type of the target interaction type.

[0026] In an embodiment of the present invention, exemplarily, before the server prepares to infer the matching type of the target interaction type, it first determines a target type matching level. This level can be a preset threshold or a dynamic value automatically adjusted by the system according to historical data and algorithms. For example, in this scenario, the target type matching level set by the server is "highly matching", which means the server will look for matching types highly relevant to the target interaction type. Based on the "highly matching" target type matching level, the server generates the second model guidance information. These information can be detailed feature descriptions of the target interaction type (such as "sports equipment"), analysis results of user behavior patterns, or preference degrees for specific types of advertisements. For example, the server can analyze that "User123" has a higher click-through rate for advertisements of high-end sports brands, and thus takes this as part of the second model guidance information. The server inputs the generated second model guidance information into a pre-trained target advertisement interaction recommendation model. This model can understand and process these guidance information to more accurately infer the advertisement types highly matching the target interaction type. After receiving the second model guidance information, the advertisement interaction recommendation model starts to work. It combines the user's historical data, behavior patterns, and preference degrees in the second model guidance information, etc., for complex calculations and inferences. For example, the model discovers that "User123" not only pays attention to sports equipment but also particularly favors high-end brands and products with innovative technologies. Therefore, the model will take these factors into consideration and infer the advertisement types highly matching the "sports equipment" target interaction type. Finally, the advertisement interaction recommendation model infers that the advertisement types highly matching "sports equipment" are "high-end brand sports equipment advertisements" and "sports equipment advertisements with innovative technologies". The server determines these inference results as the target inferred matching interaction types and selects and pushes the corresponding advertisement content for "User123" accordingly. Through the above steps, the server can more accurately infer the advertisement types highly matching the user's interests and behavior patterns based on them, and perform personalized advertisement pushing accordingly.

[0027] In an embodiment of the present invention, the target advertisement interaction recommendation model is obtained in the following manner.

[0028] Determine the account access data of the user account according to the interaction content interacted by the user account within a preset promotion period, and determine the interaction type matching the preset promotion period according to the interaction content included in the account access data; the interaction type includes a first interaction type and a second interaction type;

[0029] In the user account, determine the first account number of the first user account matching the first interaction type, determine the second account number of the second user account matching the second interaction type, and determine the third account number of the third user account matching both the first interaction type and the second interaction type;

[0030] Determine the matching degree of the interaction type between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number;

[0031] If the matching degree of the interaction type indicates that the second interaction type is the matching type of the first interaction type, then determine the first interaction type and the second interaction type as a binary training group for training the original advertisement interaction recommendation model;

[0032] Load the first interaction type in the binary training group into the original advertisement interaction recommendation model, infer the matching type of the first interaction type by the original advertisement interaction recommendation model, and determine the inferred matching interaction type of the first interaction type as the sample inferred matching interaction type;

[0033] Perform a cyclic optimization training process on the original advertisement interaction recommendation model according to the sample inferred matching interaction type and the second interaction type in the binary training group to obtain a target advertisement interaction recommendation model.

[0034] In an embodiment of the present invention, exemplarily, the server first analyzes the interaction content of the user within a preset promotion period (such as the past six months). For the user account "User123", the server finds that the user has visited the pages of sports shoes and sportswear multiple times and made purchases during the promotion period. Based on this data, the server determines two types of interactions: the first interaction type is "browsing and purchasing of sports shoes", and the second interaction type is "browsing and purchasing of sportswear". The server further analyzes all user accounts and finds the number of accounts related to the first interaction type and the second interaction type. For example, the server finds that 1000 user accounts match "browsing and purchasing of sports shoes" (the first account number), 800 user accounts match "browsing and purchasing of sportswear" (the second account number), and 600 user accounts match both of these two interaction types (the third account number). Using the first, second, and third account numbers, the server calculates a matching degree of the interaction types. This calculation can be based on co-occurrence frequency, Jaccard similarity, or other relevant metrics. In this example, the server finds that there is a high matching degree between "browsing and purchasing of sports shoes" and "browsing and purchasing of sportswear" because these two behaviors often occur simultaneously among users. Due to the high matching degree of the interaction types, the server determines "browsing and purchasing of sports shoes" (the first interaction type) and "browsing and purchasing of sportswear" (the second interaction type) as a binary training group for training the advertisement interaction recommendation model. The server loads the interaction type of "browsing and purchasing of sports shoes" into the original advertisement interaction recommendation model. This model is an initial version of the recommendation system and needs to be trained to optimize its recommendation accuracy. The original advertisement interaction recommendation model tries to infer its matching type based on the input first interaction type. In the initial stage, the inference of the model may not be completely consistent with the actual situation. Therefore, the server uses the actual matching type of "browsing and purchasing of sportswear" to compare with the inference result of the model, and then adjusts and optimizes the model according to the difference. This process is carried out cyclically until the inference result of the model is highly consistent with the actual matching type. After multiple rounds of cyclic optimization training, the server obtains a more accurate target advertisement interaction recommendation model. This model can now more precisely recommend advertisement content related to "browsing and purchasing of sportswear" based on the user's behavior of "browsing and purchasing of sports shoes". Through the above steps, the server uses the user's historical interaction data to train and optimize an advertisement interaction recommendation model, enabling it to more accurately recommend relevant advertisement content according to the user's current behavior.

[0035] In order to more clearly describe the solution provided by the embodiments of the present application, a more detailed description will be given below.

[0036] In an embodiment of the present invention, a user account refers to a unique account used to identify and record user information on a network platform. In network services, users log in through the account to enjoy personalized services and functions. For example, on an e-commerce website, "User123" is a user account. Users log in to the website through this account to browse products, place orders, etc. The preset promotion period refers to a specific time period set before analyzing user behavior or conducting marketing activities. This time period can be several days, weeks, months, or longer, and is specifically determined according to the needs of the analysis or marketing activities. Suppose an e-commerce platform wants to analyze user shopping behavior in spring. Then it will set a preset promotion period from March to May and collect user interaction data during this time period. Interaction content refers to various data and information generated when users interact with the network platform, including but not limited to the pages browsed, keywords searched, advertisements clicked, products purchased, etc. On an e-commerce website, user "User123" browsed the sports shoes page, searched for the keyword "running shoes recommendation", and clicked on an advertisement for a pair of sports shoes. These are all interaction content. Account access data refers to the access records and data associated with the user account, including access time, accessed page, access duration, operation behavior, etc. These data are usually used to analyze user behavior patterns and preferences. For the user account "User123", the server will record all the pages accessed, stay time, click behavior, etc. during the preset promotion period to form the access data of this account. The interaction type matching the preset promotion period refers to the type of interaction behavior with common characteristics or patterns shown by the user account during the preset promotion period. These types are determined by analyzing the interaction content in the account access data. During the preset promotion period of the spring shopping activity, the server found that user "User123" browsed and purchased sports shoes multiple times. Therefore, "sports shoes browsing and purchase" was determined as an interaction type matching this promotion period. The first interaction type and the second interaction type are specific classifications of the user interaction behaviors identified during the preset promotion period. The first interaction type and the second interaction type are two different types of interaction behaviors divided according to the characteristics and frequencies of the interaction content in the account access data. When analyzing the account access data of user "User123", the server found that this user both browsed and purchased sports shoes (the first interaction type) and browsed and purchased sportswear (the second interaction type). Because these two behaviors are frequent and related, they are respectively defined as the first interaction type and the second interaction type.

[0037] In addition, the first user account refers to the set of user accounts that match the first interaction type, and the second user account refers to the set of user accounts that match the second interaction type. In other words, these accounts have respectively shown behaviors that conform to the first interaction type or the second interaction type during the preset promotion period. If the first interaction type is "browsing sports shoes", then all user accounts that have browsed sports shoes during the preset promotion period will be classified as the first user accounts. Similarly, if the second interaction type is "purchasing sports accessories", then all user accounts that have purchased sports accessories during the preset promotion period will be classified as the second user accounts. The first account number refers to the number of the first user accounts that match the first interaction type. The second account number refers to the number of the second user accounts that match the second interaction type. The third account number refers to the number of the third user accounts that match both the first interaction type and the second interaction type. These numbers are used to quantify the scale of user groups with different types of interaction behaviors, and then analyze the relevance and trends between user behaviors. Suppose that during the preset promotion period, 1000 user accounts have browsed sports shoes (the first interaction type), then the first account number is 1000. At the same time, 800 user accounts have purchased sports accessories (the second interaction type), then the second account number is 800. If 600 of these user accounts have both browsed sports shoes and purchased sports accessories, then the third account number is 600. Through the statistics and analysis of these numbers, it can help the server to understand the user behavior patterns more deeply, optimize the advertising recommendation strategy, and improve the accuracy of advertising and user satisfaction. For example, if it is found that the proportion of the third account number is relatively high, indicating a strong correlation between the behaviors of browsing sports shoes and purchasing sports accessories, then when recommending advertisements, the combination recommendation of these two types of products or services can be considered.

[0038] In addition, the interaction type matching degree is an indicator to measure the strength of the correlation between the first interaction type and the second interaction type. It is usually calculated based on the number of the first accounts, the number of the second accounts, and the number of the third accounts. A high matching degree means that the two interaction types often occur together, indicating that users are more inclined to perform the second interaction after the first interaction, or there is some internal connection between the two interactions. Using the previous example, if 500 users have both browsed and purchased e-books, accounting for half of the users who browsed e-books (1000) and most of the users who purchased e-books (800), then it can be considered that the matching degree between the two interaction types of "browsing e-books" and "purchasing e-books" is relatively high. The server will collect the interaction data of users within the preset promotion period and count the number of the first accounts, the number of the second accounts, and the number of the third accounts. Then, the server will use these numbers to calculate one or more indicators, such as Jaccard similarity, co-occurrence frequency, etc., to quantify the matching degree between the two interaction types. For example, the Jaccard similarity can be calculated by dividing the number of the third accounts by the union of the number of the first accounts and the number of the second accounts, and the resulting value ranges from 0 to 1. The larger the value, the higher the matching degree between the two interaction types. In this way, the server can accurately understand the relationship between different interaction types, and then optimize the recommendation algorithm to improve the user experience and business conversion rate.

[0039] In addition, a binary training group refers to a data pair composed of two highly correlated interaction types, and these data pairs will be used to train an advertisement interaction recommendation model. In this context, the first interaction type and the second interaction type are selected as a training group because of their high matching degree. In the example of the embodiment of the present invention, "browsing sports shoes" and "purchasing sports shoes" are selected as a binary training group because of their high matching degree. This training group will be used to teach the advertisement interaction recommendation model that when the user performs the behavior of "browsing sports shoes", the model should recommend advertisements or content related to "purchasing sports shoes". The original advertisement interaction recommendation model is a model based on machine learning or deep learning, and its purpose is to recommend relevant advertisement content according to the user's interaction behavior. The model learns the user's behavior patterns through a large amount of training data (such as binary training groups) and makes recommendations based on this. On an e-commerce platform, this model will recommend advertisements or products that the user may be interested in according to the user's historical interaction behaviors (such as browsing, searching, purchasing, etc.). By using binary training groups such as "browsing sports shoes" and "purchasing sports shoes" to train the model, the accuracy of the model in recommending sports shoe-related advertisements can be improved. After receiving the first interaction type, the model will infer the second interaction type that best matches the first interaction type based on the behavior patterns and relationships it has learned. This is a prediction process, and the model will try to find the subsequent behavior or interest point that is most relevant to the given behavior. After receiving the user's behavior of "browsing electronic products", the model will analyze the historical data and find the subsequent behavior that is most relevant to this behavior, which may be "purchasing electronic products" or "purchasing electronic product accessories", etc. The second interaction type that the model infers as the best match for the first interaction type is determined as the sample inference matching interaction type of the first interaction type. This is the prediction result of the model based on the training data. If the model infers that the second interaction type that best matches the behavior of "browsing electronic products" is "purchasing electronic product accessories", then "purchasing electronic product accessories" is determined as the sample inference matching interaction type of the behavior of "browsing electronic products".

[0040] In addition, the sample inference matching interaction type refers to the matching type inferred by the original advertisement interaction recommendation model based on the first interaction type. In other words, it is the most likely subsequent interaction behavior or user interest point predicted by the model according to the input first interaction type. If the user's first interaction type is "browsing sports shoes", the model will infer "purchasing sports shoes" or "viewing sports shoe details" as the sample inference matching interaction type. In the binary training group, the second interaction type is a user behavior that is highly correlated with the first interaction type and actually occurs. This is real user behavior data, which is used to compare with the inference result of the model. In the binary training group of "browsing sports shoes" and "purchasing sports shoes", "purchasing sports shoes" is the second interaction type, which represents the behavior that the user actually performs after browsing sports shoes.

[0041] In an embodiment of the present invention, determining the first account number of the first user account that matches the first interaction type in the user account can be implemented through the following examples.

[0042] Extract from the account access data the account access data whose contained interaction content is classified as the first interaction type, and determine the extracted account access data whose contained interaction content is classified as the first interaction type as the first account access data;

[0043] In the user account, determine the user account corresponding to the first account access data as the first user account;

[0044] Determine the first account number of the first user account according to the first account access data of the first user account.

[0045] In an embodiment of the present invention, for example, the server filters out the account access data related to the first interaction type (such as "browsing electronic products") from a large amount of account access data. This data includes information such as the time when the user browses electronic products, the categories of electronic products browsed, and the stay time. The server quickly locates and extracts all the account access data classified as "browsing electronic products" through a specific data extraction algorithm, and then marks these data as the first account access data. After extracting the first account access data, the server further analyzes the user accounts corresponding to these data. By comparing the associated information between the account access data and the user accounts, the server can determine which user accounts have performed the interaction behavior of "browsing electronic products". These determined user accounts are called the first user accounts. After determining the first user accounts, the server needs to count these accounts to determine the number of user accounts that have performed the interaction behavior of "browsing electronic products". This number is called the first account number. The server traverses the list of the first user accounts and performs deduplication counting to finally obtain the accurate first account number. This number reflects how many independent user accounts have performed the specified type of interaction behavior, which is of great significance for subsequent data analysis and advertising recommendation strategy formulation. Through the above three steps, the server can accurately count the number of user accounts that have performed a specific interaction behavior, providing strong data support for further user behavior analysis and accurate advertising push.

[0046] In an embodiment of the present invention, the first user account includes multiple first reference user accounts;

[0047] Determining the first account number of the first user account according to the first account access data of the first user account can be implemented through the following examples.

[0048] Extract a target first reference user account from the multiple first reference user accounts;

[0049] Determine the first account access data of the target first reference user account as the first pending access data;

[0050] Perform a cutting process on the first pending access data according to the advertising placement channels where the interaction content included in the first pending access data is located, to obtain multiple first access data sets; one first access data set includes at least one first pending access data, and the advertising placement channels where the interaction content included in the first pending access data in one first access data set are the same; the advertising placement channel where the interaction content included in the first pending access data in one first access data set is the advertising placement channel corresponding to one first access data set; the advertising placement channels corresponding to the multiple first access data sets are different;

[0051] Determine the number of accounts of the target first reference user account corresponding to the first interaction type as the number of the multiple first access data sets;

[0052] Starting from the first first reference user account, obtain the number of accounts of the multiple first reference user accounts corresponding to the first interaction type respectively;

[0053] Determine the first account number of the first user account according to the number of accounts of the multiple first reference user accounts corresponding to the first interaction type.

[0054] In an embodiment of the present invention, exemplarily, the server stores a large amount of user account data, including multiple accounts marked as first reference user accounts. These first reference user accounts are selected by the server according to certain rules (such as activity level, interaction frequency, etc.). Now, the server needs to extract a target first reference user account from these first reference user accounts for processing. The server successfully extracts a target first reference user account according to a preset extraction rule (such as random extraction, extraction after sorting by account activity level, etc.). After extracting the target first reference user account, the server further obtains the first account access data of this account. These data record all the interaction behaviors of the user account within a specific time period. The server marks these data as first pending access data and prepares for subsequent processing. The server performs cutting processing on these data according to the advertising placement channels where the interaction content included in the first pending access data is located. For example, some interaction content is generated under the social media advertising channel, while some other interaction content can be generated under the search engine advertising channel. The server classifies these data according to the advertising placement channels to obtain multiple first access data sets. The interaction content in each set comes from the same advertising placement channel. After the cutting processing, the server obtains multiple first access data sets. Each set represents the interaction behavior of the user under a specific advertising placement channel. The server determines the number of these sets as the account number corresponding to the first interaction type (such as "click on an advertisement") of the target first reference user account. This number reflects the number of responses of the user to the first interaction type under different advertising placement channels. The server follows the above steps and processes all the first reference user accounts one by one starting from the first first reference user account. For each account, the server extracts its first pending access data, performs cutting processing, and determines the account number corresponding to the first interaction type. After processing all the first reference user accounts, the server performs summarization and analysis according to the account numbers corresponding to the first interaction type of these accounts. The server can use statistical methods such as summation, averaging, or other methods to synthesize these numbers and finally determine the overall response situation of the first user account in the first interaction type, that is, the first account number. This number provides a quantitative indicator for the server regarding the overall activity level and response situation of the user in a specific interaction type.

[0055] In an embodiment of the present invention, the determination of the second account number of the second user account matching the second interaction type can be implemented through the following examples.

[0056] Extract the account access data whose included interaction content is classified as the second interaction type from the account access data, and determine the account access data whose included interaction content is classified as the second interaction type and is extracted as the second account access data;

[0057] In the user account, determine the user account corresponding to the second account access data as the second user account;

[0058] Determine the second account number of the second user account according to the second account access data of the second user account.

[0059] In an embodiment of the present invention, exemplarily, the server has a large amount of account access data, which records various interaction behaviors of users. Now, the server needs to determine the number of user accounts matching the second interaction type (such as "purchasing electronic products"). First, the server runs a data screening algorithm to extract all interaction records classified as "purchasing electronic products" from the massive account access data. These records constitute the second account access data, which may include information such as purchase time, details of the purchased products, transaction status, etc. After extracting the second account access data, the server further analyzes these data to determine which user accounts have performed the interaction behavior of "purchasing electronic products". By comparing the user identifiers in the account access data with the user account database, the server can accurately identify the user accounts that have made purchases, and these accounts are marked as the second user accounts. Finally, the server needs to count the second user accounts determined in the second step to obtain the total number of user accounts that have performed the interaction behavior of "purchasing electronic products". The server traverses the list of second user accounts and performs deduplication to ensure that each account is counted only once. Finally, the server will obtain a specific number, that is, the second account number, which represents how many independent user accounts have completed the interaction behavior of "purchasing electronic products". This information is crucial for evaluating product sales, analyzing user purchase behavior, and formulating future marketing strategies. Through these three steps, the server can accurately count the number of user accounts matching a specific interaction type (in this example, "purchasing electronic products"), thereby providing strong support for the enterprise's data analysis and business decisions.

[0060] In an embodiment of the present invention, the account access data includes the first account access data of the first user account and the second account access data of the second user account; the first account access data refers to the account access data whose included interaction content is classified as the first interaction type; the second account access data refers to the account access data whose included interaction content is classified as the second interaction type;

[0061] The determination of the third account number of the third user account that simultaneously matches the first interaction type and the second interaction type can be implemented through the following example.

[0062] From the first user account, determine the first user account classified as the second user account, and determine the first user account classified as the second user account as the third user account;

[0063] Determine the third account number of the third user account according to the first account access data of the third user account and the second account access data of the third user account.

[0064] In an embodiment of the present invention, exemplarily, the server stores a large amount of account access data, including the first account access data of the first user account and the second account access data of the second user account. The first account access data refers to the account access data whose included interaction content is classified as the first interaction type (such as "browsing electronic products"), while the second account access data refers to the account access data whose included interaction content is classified as the second interaction type (such as "purchasing electronic products"). Now, the server needs to determine from the first user account those accounts that also belong to the second user account, that is, the user accounts that have both "browsed electronic products" and "purchased electronic products". The server finds the intersection of the first user account and the second user account by comparing their lists, and the user accounts in these intersections are the third user accounts. After determining the third user accounts, the server needs to further determine the third account number based on the first account access data and the second account access data of these accounts. The third account number here does not simply calculate the number of third user accounts, but involves a comprehensive consideration of the behavior frequencies, activity levels, etc. of these accounts in the two interaction types. For example, the server will analyze data such as the number of times and time intervals of each third user account browsing and purchasing electronic products, and then give a comprehensive score or weight based on these data. Finally, the server will summarize these scores or weights to obtain a third account number that reflects the comprehensive activity level of the third user account in the two interaction types. Suppose the server finds through analysis that 100 user accounts have both browsed and purchased electronic products, that is, these 100 accounts are the third user accounts. Further analyzing the access data of these accounts, the server finds that some of these accounts are more frequent in browsing and purchasing behaviors, so higher scores will be given to these accounts. Finally, the server can calculate a weighted third account number based on these scores, such as 85 (this number is assumed for illustrative purposes), which is neither exactly equal to 100 (because the activity levels of all accounts are not the same) nor just a simple account number statistic, but reflects the comprehensive activity level of these accounts in the two interaction behaviors. In summary, through these two steps, the server can accurately identify the user accounts that have both performed the first interaction type and the second interaction type, and give a comprehensive third account number based on the access data of these accounts, which is of great significance for analyzing user behavior, evaluating marketing effects, and optimizing product recommendation strategies.

[0065] In an embodiment of the present invention, the third user account includes a plurality of intermediate user accounts;

[0066] The determination of the third account number of the third user account according to the first account access data of the third user account and the second account access data of the third user account can be executed and implemented through the following examples.

[0067] Extract a target intermediate user account from the multiple intermediate user accounts;

[0068] Determine the number of accounts corresponding to the first interaction type and the second interaction type of the target intermediate user account according to the first account access data of the target intermediate user account and the second account access data of the target intermediate user account;

[0069] Starting from the first intermediate user account, obtain the number of accounts corresponding to the first interaction type and the second interaction type of the multiple intermediate user accounts respectively;

[0070] Determine the third account number of the third user account according to the number of accounts corresponding to the first interaction type and the second interaction type of the multiple intermediate user accounts.

[0071] In an embodiment of the present invention, exemplarily, a large amount of third user account data is stored in the server, and these accounts are further subdivided into multiple intermediate user accounts. Each intermediate user account contains detailed information about the user's interaction with the platform. For more in-depth analysis, the server needs to extract a target account from these intermediate user accounts for processing. For example, the server can select a target intermediate user account based on the account's activity level, interaction frequency, or pattern of specific behaviors. Suppose the server selects an intermediate user account that has both browsing and purchasing behaviors in the past week as the target account. After extracting the target intermediate user account, the server analyzes the first account access data and the second account access data of this account. These data record the user's behaviors under different interaction types, such as browsing and purchasing. The server will first count the number of times the target account browses electronic products (the first interaction type), for example, it has browsed 10 times in the past week. Then, the server will count the number of times the account purchases electronic products (the second interaction type), for example, it has purchased 2 times in the same week. These two numbers represent the corresponding account numbers of the target intermediate user account in the first interaction type and the second interaction type. The server will follow the above method and analyze all intermediate user accounts one by one starting from the first intermediate user account. For each account, the server will count the number of times of its behaviors under the first interaction type (browsing) and the second interaction type (purchasing). This process involves a large amount of data processing and calculation, but finally the server will obtain a dataset containing the number of times of behaviors of each intermediate user account under the two interaction types. After analyzing all intermediate user accounts, the server will determine the third account number of the third user account based on the collected data. This number may not be just a simple sum or average, but the result obtained based on a more complex algorithm or model. For example, the server will consider factors such as the frequency of account behaviors, time distribution, and the correlation between behaviors, and use machine learning algorithms to predict or evaluate the overall activity of the third user account in the two interaction types. The finally obtained third account number can be a weighted metric, reflecting the overall performance of these accounts in the two interaction behaviors. Through these steps, the server can understand the user's behavior patterns in the two interaction types more deeply, providing valuable insights for the enterprise's marketing strategies, product recommendation systems, etc.

[0072] In an embodiment of the present invention, the determining of the corresponding account numbers of the target intermediate user account in the first interaction type and the second interaction type according to the first account access data of the target intermediate user account and the second account access data of the target intermediate user account can be implemented through the following examples.

[0073] Determine the first account access data of the target intermediate user account as the first intermediate access data, and determine the second account access data of the target intermediate user account as the second intermediate access data;

[0074] According to the advertising placement channels where the interaction content included in the first intermediate access data is located, perform cutting processing on the first intermediate access data to obtain multiple first intermediate access data sets; one first intermediate access data set includes at least one first intermediate access data, and the advertising placement channels where the interaction content included in the first intermediate access data in one first intermediate access data set are the same; the advertising placement channel where the interaction content included in the first intermediate access data in one first intermediate access data set is the advertising placement channel corresponding to one first intermediate access data set; the advertising placement channels corresponding to the multiple first intermediate access data sets are different;

[0075] According to the advertising placement channels where the interaction content included in the second intermediate access data is located, perform cutting processing on the second intermediate access data to obtain multiple second intermediate access data sets; one second intermediate access data set includes at least one second intermediate access data, and the advertising placement channels included in the second intermediate access data in one second intermediate access data set are the same; the advertising placement channel included in the second intermediate access data in one second intermediate access data set is the advertising placement channel corresponding to one second intermediate access data set; the advertising placement channels corresponding to the multiple second intermediate access data sets are different;

[0076] According to the multiple first intermediate access data sets and the multiple second intermediate access data sets, determine the number of accounts of the target intermediate user account corresponding to the first interaction type and the second interaction type.

[0077] In an embodiment of the present invention, exemplarily, the server first clearly distinguishes the first account access data and the second account access data from the account access data of the target intermediate user account. The first account access data mainly records the behavior of the user browsing electronic products, while the second account access data records the behavior of the user purchasing electronic products. The server marks these data as the first intermediate access data and the second intermediate access data respectively for subsequent processing. Next, the server analyzes the first intermediate access data, especially the advertising channels where the interaction content contained therein is located. For example, the user may have clicked and browsed electronic products through different channels such as search engine advertisements, social media advertisements, or email marketing. The server cuts the first intermediate access data according to these advertising channels to form multiple first intermediate access data sets. The data in each set comes from the same advertising channel. Similar to the processing method of the first intermediate access data, the server also performs cutting processing on the second intermediate access data. This time, it is based on the advertising channels where the user purchases electronic products. For example, the user may have made a purchase through channels such as recommended advertisements on online shopping platforms, television shopping advertisements, or promotional advertisements in offline physical stores. The server cuts the second intermediate access data into multiple second intermediate access data sets according to these different advertising channels. After completing the data cutting process, the server will have multiple first intermediate access data sets and multiple second intermediate access data sets. Next, the server analyzes these sets to determine how many times the target intermediate user account has browsed (the first interaction type) and purchased (the second interaction type) under different advertising channels. For example, the server finds that through the search engine advertising channel, the user has browsed electronic products 5 times and purchased 2 times; while through the social media advertising channel, the user has browsed 3 times but has not made a purchase. These data will provide the server with in-depth insights into the user's behavior patterns under different advertising channels, thereby helping enterprises to more accurately place advertisements and formulate marketing strategies. Through these steps, the server can accurately understand the browsing and purchasing behaviors of the target intermediate user account under different advertising channels, and provide more effective marketing and advertising placement suggestions for enterprises.

[0078] In an embodiment of the present invention, according to the multiple first intermediate access data sets and the multiple second intermediate access data sets, determining the number of accounts corresponding to the first interaction type and the second interaction type of the target intermediate user account can be implemented through the following examples.

[0079] Construct at least one binary intermediate access data group according to the multiple first intermediate access data sets and the multiple second intermediate access data sets; any binary intermediate access data group includes a first intermediate access data set and a second intermediate access data set;

[0080] Determine the number of accounts corresponding to the first interaction type and the second interaction type for the target intermediate user account according to the at least one binary intermediate access data group.

[0081] In an embodiment of the present invention, by way of example, the server has previously performed cutting processing on the first intermediate access data and the second intermediate access data of the target intermediate user account to obtain a plurality of first intermediate access data sets and a plurality of second intermediate access data sets. Now, the server needs to pair these sets to construct at least one binary intermediate access data group. For example, the server may have a first intermediate access data set A that records data on users browsing electronic products through the search engine advertising channel. At the same time, the server also has a second intermediate access data set B that records data on users purchasing electronic products through the same search engine advertising channel. The server will pair these two sets to form a binary intermediate access data group (A, B). The server will construct multiple such binary intermediate access data groups, each of which represents the browsing and purchasing behaviors of users under a specific advertising placement channel. After constructing the binary intermediate access data groups, the server will analyze the data in these groups to determine the number of accounts of the target intermediate user account under the first interaction type (browsing electronic products) and the second interaction type (purchasing electronic products). Taking the binary intermediate access data group (A, B) as an example, the server will count the number of times users browse electronic products in set A and the number of times users purchase electronic products in set B. For example, in set A, users browse electronic products 10 times through the search engine advertising; in set B, users purchase electronic products 2 times through the same channel. The server will determine, based on this data, that the number of times the target intermediate user account browses electronic products through the search engine advertising channel under the first interaction type is 10 times; and the number of times of purchasing electronic products under the second interaction type is 2 times. These specific numbers represent the number of accounts of the target intermediate user account under the two interaction types. Through these steps, the server can accurately analyze the number of browsing and purchasing behaviors of the target intermediate user account under different advertising placement channels, thereby providing more detailed market analysis and user behavior insights for enterprises.

[0082] In an embodiment of the present invention, the determining of the interaction type matching degree between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number can be implemented through the following example.

[0083] Perform a number statistics on the third account number according to the interaction occurrence time information of the interaction content included in the account access data to obtain the target account number of the third user account;

[0084] The degree of interaction type matching between the first interaction type and the second interaction type is determined according to the number of the first accounts, the number of the second accounts, and the number of the target accounts.

[0085] In an embodiment of the present invention, the server first obtains account access data, which contains detailed records of the user's interaction with the platform, and an important piece of information is the time when the interaction occurs. The server pays special attention to the behavior records of third user accounts (these accounts involve both the first interaction type and the second interaction type). For example, the server finds that in a certain time period (such as the past week), the third user account performs the second interaction type (such as purchasing electronic products) immediately after performing the first interaction type (such as browsing electronic products). The server will count how many third user accounts have shown this conversion behavior from browsing to purchasing in this time period based on these time information. This statistical result is called the number of target accounts, which reflects the number of third user accounts that have actually completed the conversion from the first interaction type to the second interaction type. After obtaining the number of first accounts (the number of accounts that only perform the first interaction type), the number of second accounts (the number of accounts that only perform the second interaction type) and the number of target accounts (the number of accounts that perform both interaction types at the same time), the server will perform further analysis. The server will use an algorithm or model, such as association rule mining or conversion rate calculation, to evaluate the degree of match between the first interaction type and the second interaction type. Specifically, the server will look at the ratio of the number of target accounts to the sum of the number of first accounts and the number of second accounts. This ratio can reflect the correlation or conversion efficiency between the two interaction types to a certain extent. For example, if the server finds that a high proportion of users who browsed electronic products eventually made a purchase, then it can be considered that the first interaction type (browsing electronic products) and the second interaction type (purchasing electronic products) have a high degree of match. Conversely, if the conversion rate is very low, the match is low. Through such analysis, the server can provide companies with valuable market insights, help companies understand user behavior patterns, optimize marketing strategies, and improve user conversion rates and satisfaction.

[0086] In the embodiment of the present invention, determining the degree of interaction type matching between the first interaction type and the second interaction type according to the first account number, the second account number and the third account number can be implemented through the following examples.

[0087] determining a first type matching degree between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number;

[0088] Determine the interaction pattern consistency between the first interaction type and the second interaction type, and determine the interaction pattern consistency as the matching degree of the second type between the first interaction type and the second interaction type;

[0089] Determine the matching degree of the interaction types between the first interaction type and the second interaction type according to the first type matching degree and the second type matching degree.

[0090] In an embodiment of the present invention, exemplarily, the server will first collect the number of first accounts (the number of accounts only participating in the first interaction type), the number of second accounts (the number of accounts only participating in the second interaction type), and the number of third accounts (the number of accounts participating in both interaction types). Based on these data, the server will calculate the first type matching degree between the first interaction type and the second interaction type. For example, on an e-commerce platform, the first interaction type can be that a user browses products, and the second interaction type can be that a user purchases products. The server statistics show that there are 1000 accounts that only browsed products (the number of first accounts), 200 accounts that only purchased products (the number of second accounts), and 300 accounts that both browsed and purchased products (the number of third accounts). By comparing these numbers, the server can calculate the conversion rate from browsing to purchasing, that is, the ratio of the number of third accounts to the number of first accounts, and this ratio can be used as an indicator of the first type matching degree. In this example, the conversion rate is 30% (300 / 1000), indicating that 30% of the users who browsed products finally made a purchase. Next, the server will analyze the consistency of the interaction patterns between the first interaction type and the second interaction type. This involves whether the behavior patterns and sequences of users when performing the two interaction types are consistent. The server will observe the users who both browsed and purchased products (the users in the number of third accounts) and analyze their behavior paths. If most users first browse products and then make a purchase, this indicates that there is a sequential consistency between the two interaction types, that is, users usually make a purchase after understanding the product details. The server will determine the second type matching degree according to this degree of consistency. For example, if 90% of the users in the number of third accounts follow the sequence of browsing first and then purchasing, then the second type matching degree can be considered high. Finally, the server will synthesize the first type matching degree and the second type matching degree to determine the overall interaction type matching degree between the first interaction type and the second interaction type. In the above example of the e-commerce platform, the server has calculated that the first type matching degree is 30% (conversion rate), and the second type matching degree is 90% (interaction pattern consistency). The server will use a weighted average or other algorithms to combine these two indicators to obtain an overall interaction type matching degree. For example, if the server gives the same weight to the two indicators, then the overall interaction type matching degree can be the average of the two, that is, 60%. This value reflects the overall matching and conversion efficiency between the two interaction types of browsing products and purchasing products.

[0091] In an embodiment of the present invention, the interaction actions of the user account on the interaction content include a first interaction action and a second interaction action; the first account number includes: a first intermediate account number matching the first interaction action and a second intermediate account number matching the second interaction action; the second account number includes: a third intermediate account number matching the first interaction action and a fourth intermediate account number matching the second interaction action; the third account number includes: a fifth intermediate account number matching the first interaction action and a sixth intermediate account number matching the second interaction action;

[0092] Determining the matching degree of the interaction types between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number can be implemented through the following examples.

[0093] Determine the matching degree of the interaction types between the first interaction type and the second interaction type corresponding to the first interaction action according to the first intermediate account number, the third intermediate account number, and the fifth intermediate account number;

[0094] Determine the matching degree of the interaction types between the first interaction type and the second interaction type corresponding to the second interaction action according to the second intermediate account number, the fourth intermediate account number, and the sixth intermediate account number;

[0095] Determine the matching degree of the interaction types between the first interaction type and the second interaction type according to the matching degree of the interaction types corresponding to the first interaction action and the matching degree of the interaction types corresponding to the second interaction action.

[0096] In an embodiment of the present invention, exemplarily, on an online video platform, the server records the interaction behaviors of users with videos, including two interaction actions: liking (the first interaction action) and commenting (the second interaction action). The server wants to analyze whether users are more inclined to purchase a platform membership (the second interaction type) after watching a certain type of video (the first interaction type). For the interaction action of liking, the server counts the following account numbers: The first intermediate account number: the number of accounts that have watched a certain type of video and liked it. The third intermediate account number: the number of accounts that have purchased a platform membership and liked it. The fifth intermediate account number: the number of accounts that have both watched a certain type of video and purchased a platform membership and liked it. By comparing these account numbers, especially the relationship between the fifth intermediate account number and the first and third intermediate account numbers, the server can calculate the proportion of users who have purchased a platform membership among the users who liked a certain type of video. This proportion is the type matching degree between the first interaction type and the second interaction type for the interaction action of liking. Continuing with the example of the above online video platform, for the interaction action of commenting, the server also counts the following account numbers: The second intermediate account number: the number of accounts that have watched a certain type of video and commented. The fourth intermediate account number: the number of accounts that have purchased a platform membership and commented. The sixth intermediate account number: the number of accounts that have both watched a certain type of video and purchased a platform membership and commented. Through a similar analysis method, the server can calculate the proportion of users who have purchased a platform membership among the users who commented after watching a certain type of video. This proportion reflects the type matching degree between the first interaction type and the second interaction type for the interaction action of commenting. After obtaining the type matching degrees for the two interaction actions of liking and commenting, the server will comprehensively consider these two metrics to determine the overall type matching degree between watching a certain type of video and purchasing a platform membership. The server will use algorithms such as weighted average, taking the maximum value, or other appropriate algorithms to combine the matching degrees of these two interaction actions. For example, if the server believes that liking and commenting are equally important, then the average of the two can be used as the overall type matching degree. This overall matching degree will more comprehensively reflect the correlation and conversion efficiency between watching a certain type of video and purchasing a platform membership.

[0097] In an embodiment of the present invention, the type matching degree is used to indicate the type matching level of the second interaction type with respect to the first interaction type;

[0098] Loading the first interaction type in the binary training group into the original advertisement interaction recommendation model, inferring the matching type of the first interaction type by the original advertisement interaction recommendation model, and determining the inferred matching interaction type of the first interaction type as the sample inferred matching interaction type of the first interaction type can be implemented through the following example.

[0099] Generate first model guidance information matching the first interaction type according to the type matching level;

[0100] Load the first model guidance information into the original advertisement interaction recommendation model. The original advertisement interaction recommendation model infers, according to the first model guidance information, a matching type with a matching level of the type matching level between the first interaction type, and determines the matching type with a matching level of the type matching level between the first interaction type as the sample inferred matching interaction type of the first interaction type.

[0101] In an embodiment of the present invention, by way of example, on an online shopping platform, the server determines the interaction type matching degree between browsing a certain type of commodity (the first interaction type) and purchasing the certain type of commodity (the second interaction type) by analyzing the user's browsing and purchasing behaviors. This matching degree is a specific value, such as 0.8 (ranging from 0 to 1), indicating that when a user browses a certain type of commodity, there is an 80% possibility of purchasing the certain type of commodity. The server sets the type matching level to "high matching" according to this interaction type matching degree. This means that when a user browses this type of commodity, the server can highly recommend relevant purchase options. Based on the above "high matching" level, the server generates first model guidance information matching browsing a certain type of commodity (the first interaction type). This guidance information may include a series of rules, weights or feature vectors for guiding the advertisement interaction recommendation model on how to better infer purchase recommendations matching the browsing behavior. For example, the guidance information may include "when a user browses electronic products, increase the recommendation weight for purchasing electronic products", or "for users who have browsed electronic products, preferentially recommend electronic products with high evaluations and good sales volumes". The server loads the generated first model guidance information into the original advertisement interaction recommendation model. This model may originally only recommend advertisements based on information such as the user's browsing history and purchase records, but now with the additional guidance information, the model can more accurately infer purchase recommendations matching the user's browsing behavior. When a user browses electronic products again, the advertisement interaction recommendation model will infer purchase recommendations matching the user's current browsing behavior according to the loaded first model guidance information. This inference result is the sample inferred matching interaction type, that is, the purchase options that the model believes the user may be interested in. For example, if a user is browsing a smart phone, the model will infer that "the user may be interested in purchasing a new smart phone or related accessories" and generate corresponding advertisement recommendations accordingly. This inference process is the result of the combined action of the interaction type matching degree and the first model guidance information.

[0102] In an embodiment of the present invention, the process of circularly optimizing and training the original advertisement interaction recommendation model according to the interaction type inferred and matched from the sample and the second interaction type in the binary training group to obtain a target advertisement interaction recommendation model can be implemented through the following examples.

[0103] Infer and match the interaction type from the sample and the second interaction type, and determine a cost parameter for the original advertisement interaction recommendation model;

[0104] Perform a circular optimization training process on the original advertisement interaction recommendation model according to the cost parameter to obtain the target advertisement interaction recommendation model.

[0105] The server has obtained the interaction type inferred and matched from the sample (i.e., the interaction type that the model infers the user may be interested in) and the second interaction type in the binary training group (i.e., the actual user interaction behavior) through the previous steps. Now, the server needs to compare the differences between the two to determine the cost parameter for the original advertisement interaction recommendation model. For example, in an online video recommendation system, the interaction type inferred and matched from the sample can be "the user may be interested in science fiction movies", while the second interaction type can be "the user actually watched a comedy movie". The server will calculate the difference between the two, such as by comparing their feature vectors, user ratings, or other relevant metrics. This difference value is used as the cost parameter, which reflects the deviation between the model inference and the actual user behavior. Once the cost parameter is determined, the server will start a circular optimization training process on the original advertisement interaction recommendation model according to this parameter. This process aims to minimize the cost parameter by adjusting the parameters and structure of the model, that is, to reduce the deviation between the model inference and the actual user behavior. In the circular optimization training process, the server will repeatedly perform the following steps: use the training data (including the user's historical interaction data, the interaction type inferred and matched from the sample, and the actual second interaction type) to train the model. Evaluate the performance of the model, and calculate the cost parameter by comparing the inference result of the model with the actual user behavior. Adjust the parameters and structure of the model according to the cost parameter, such as weights, biases, or the number of network layers, etc. Repeat the above steps until the preset number of training rounds is reached, the cost parameter converges, or the model performance reaches the preset standard. Through this process, the server can gradually optimize the advertisement interaction recommendation model to make it more accurately infer the interaction type that matches the user behavior. The finally obtained model is the target advertisement interaction recommendation model, which can better meet the user's needs and improve the accuracy of advertisement recommendation and user satisfaction.

[0106] An embodiment of the present invention provides a computer device 100. The computer device 100 includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned advertising analysis method based on big data. As Figure 2 shown, Figure 2 is a block diagram of the computer device 100 provided by an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To achieve data transmission or interaction, the elements of the memory 111, the processor 112, and the communication unit 113 are electrically connected to each other directly or indirectly. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines.

[0107] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. The embodiments were chosen and described in order to best illustrate the principles of the disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to adapt various embodiments with different modifications to suit the particular applications contemplated.

Claims

1. An advertising analysis method based on big data, characterized in that, The method includes: When an advertisement push instruction for a target user account is obtained, obtaining the interaction content interacted by the target user account during a past monitoring period; Determining a target interaction type matching the target user account according to the interaction content interacted by the target user account during the past monitoring period; Loading the target interaction type into a pre-trained target advertisement interaction recommendation model, inferring the matching type of the target interaction type by the target advertisement interaction recommendation model, and determining the matching type of the target interaction type inferred and determined as the target inferred matching interaction type of the target interaction type; Determining the interaction content to be pushed to the target user account from the interaction content corresponding to the target inferred matching interaction type; The target advertisement interaction recommendation model is obtained by the following method, including: Determining the account access data of the user account according to the interaction content interacted by the user account during a preset promotion period, and determining the interaction type matching the preset promotion period according to the interaction content included in the account access data; the interaction type includes a first interaction type and a second interaction type; Extracting, from the account access data, the account access data including the interaction content classified as the first interaction type, and determining the account access data including the interaction content classified as the first interaction type extracted as the first account access data; In the user account, determining the user account corresponding to the first account access data as the first user account; the first user account includes a plurality of first reference user accounts; Extracting a target first reference user account from the plurality of first reference user accounts; Determining the first account access data of the target first reference user account as the first pending access data; Performing a cutting process on the first pending access data according to the advertisement placement channel where the interaction content included in the first pending access data is located, to obtain a plurality of first access data sets; one first access data set includes at least one first pending access data, and the advertisement placement channels where the interaction content included in the first pending access data in one first access data set is located are the same; the advertisement placement channel where the interaction content included in the first pending access data in one first access data set is located is the advertisement placement channel corresponding to one first access data set; the advertisement placement channels corresponding to the plurality of first access data sets are different; Determining the number of the plurality of first access data sets as the number of accounts of the target first reference user account corresponding to the first interaction type; Starting from the first first reference user account, respectively obtaining the number of accounts of the plurality of first reference user accounts corresponding to the first interaction type; Determine the first account number of the first user account according to the number of accounts corresponding to the first interaction type among the multiple first reference user accounts. Extract the account access data containing interaction content classified as the second interaction type from the account access data, and determine the account access data containing the extracted interaction content classified as the second interaction type as the second account access data; Among the user accounts, determine the user account corresponding to the second account access data as the second user account; Determine the second account number of the second user account according to the second account access data of the second user account, and determine the third account number of the third user account that matches both the first interaction type and the second interaction type; Determine the matching degree of the interaction types between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number; the matching degree of the interaction types is used to indicate the type matching level of the second interaction type for the first interaction type; If the matching degree of the interaction types indicates that the second interaction type is the matching type of the first interaction type, then determine the first interaction type and the second interaction type as the binary training group for training the original advertisement interaction recommendation model; Generate the first model guidance information matching the first interaction type according to the type matching level; Load the first model guidance information into the original advertisement interaction recommendation model. The original advertisement interaction recommendation model infers the matching type with the matching level of the type matching level for the first interaction type according to the first model guidance information, and determine the matching type with the matching level of the type matching level for the first interaction type inferred and determined as the sample inferred matching interaction type of the first interaction type; Determine the cost parameter for the original advertisement interaction recommendation model according to the sample inferred matching interaction type and the second interaction type; Perform a cyclic optimization training process on the original advertisement interaction recommendation model according to the cost parameter to obtain the target advertisement interaction recommendation model.

2. The method according to claim 1, wherein The step of loading the target interaction type into the pre-trained target advertisement interaction recommendation model, where the target advertisement interaction recommendation model infers the matching type of the target interaction type, and determine the inferred matching type of the target interaction type as the target inferred matching interaction type of the target interaction type, includes: Obtain the target type matching level for inferring the target inferred matching interaction type; Generate the second model guidance information matching the target interaction type according to the target type matching level; Load the second model guidance information into the target advertisement interaction recommendation model. The target advertisement interaction recommendation model infers, according to the second model guidance information, a matching type whose matching level with the target interaction type is the target type matching level, and determines the matching type whose matching level with the target interaction type is the target type matching level as the target inferred matching interaction type of the target interaction type.

3. The method according to claim 1, characterized in that, The account access data includes the first account access data of the first user account and the second account access data of the second user account; the first account access data refers to the account access data whose included interaction content is classified as the first interaction type; the second account access data refers to the account access data whose included interaction content is classified as the second interaction type. Determining the number of third accounts of the third user account that matches both the first interaction type and the second interaction type includes: Determine, from the first user accounts, the first user accounts classified as the second user account, and determine the first user accounts classified as the second user account as the third user account. Determine the number of third accounts of the third user account according to the first account access data of the third user account and the second account access data of the third user account.

4. The method according to claim 3, wherein The third user account includes multiple intermediate user accounts. Determining the number of third accounts of the third user account according to the first account access data of the third user account and the second account access data of the third user account includes: Extract a target intermediate user account from the multiple intermediate user accounts. According to the first account access data of the target intermediate user account and the second account access data of the target intermediate user account, determine the number of accounts of the target intermediate user account corresponding to the first interaction type and the second interaction type. Starting from the first intermediate user account, obtain the number of accounts of the multiple intermediate user accounts corresponding to the first interaction type and the second interaction type respectively. Determine the number of third accounts of the third user account according to the number of accounts of the multiple intermediate user accounts corresponding to the first interaction type and the second interaction type.

5. The method according to claim 4, wherein Determining the number of accounts of the target intermediate user account corresponding to the first interaction type and the second interaction type according to the first account access data of the target intermediate user account and the second account access data of the target intermediate user account includes: Determine the first account access data of the target intermediate user account as the first intermediate access data, and determine the second account access data of the target intermediate user account as the second intermediate access data. According to the advertising placement channels where the interaction content in the first intermediate access data is located, perform a cutting process on the first intermediate access data to obtain multiple first intermediate access data sets; one first intermediate access data set includes at least one first intermediate access data, and the advertising placement channels where the interaction content in the first intermediate access data in one first intermediate access data set is located are the same; the advertising placement channel where the interaction content in the first intermediate access data in one first intermediate access data set is located is the advertising placement channel corresponding to one first intermediate access data set; the advertising placement channels corresponding to the multiple first intermediate access data sets are different; According to the advertising placement channels where the interaction content in the second intermediate access data is located, perform a cutting process on the second intermediate access data to obtain multiple second intermediate access data sets; one second intermediate access data set includes at least one second intermediate access data, and the advertising placement channels included in the second intermediate access data in one second intermediate access data set are the same; the advertising placement channel included in the second intermediate access data in one second intermediate access data set is the advertising placement channel corresponding to one second intermediate access data set; the advertising placement channels corresponding to the multiple second intermediate access data sets are different; According to the multiple first intermediate access data sets and the multiple second intermediate access data sets, construct at least one binary intermediate access data group; any one binary intermediate access data group includes a first intermediate access data set and a second intermediate access data set; According to the at least one binary intermediate access data group, determine the number of accounts corresponding to the target intermediate user account in the first interaction type and the second interaction type.

6. The method according to claim 1, characterized in that, The determining the matching degree of the interaction types between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number includes: According to the interaction occurrence time information of the interaction content included in the account access data, perform a number statistics on the third account number to obtain the target account number of the third user account; According to the first account number, the second account number, and the target account number, determine the matching degree of the interaction types between the first interaction type and the second interaction type.

7. The method according to claim 1, characterized in that, The determining the matching degree of the interaction types between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number includes: According to the first account number, the second account number, and the third account number, determine the first type matching degree between the first interaction type and the second interaction type; Determine the consistency of the interaction modes between the first interaction type and the second interaction type, and determine the consistency of the interaction modes as the second type matching degree between the first interaction type and the second interaction type; Determine the interaction type matching degree between the first interaction type and the second interaction type according to the first type matching degree and the second type matching degree.

8. The method according to claim 1, characterized in that The interaction actions of the user account on the interaction content include a first interaction action and a second interaction action; the first account number includes: a first intermediate account number matching the first interaction action, and a second intermediate account number matching the second interaction action; the second account number includes: a third intermediate account number matching the first interaction action, and a fourth intermediate account number matching the second interaction action; the third account number includes: a fifth intermediate account number matching the first interaction action, and a sixth intermediate account number matching the second interaction action; Determining the interaction type matching degree between the first interaction type and the second interaction type according to the first account number, the second account number, and the third account number includes: Determine the type matching degree of the first interaction type and the second interaction type corresponding to the first interaction action according to the first intermediate account number, the third intermediate account number, and the fifth intermediate account number; Determine the type matching degree of the first interaction type and the second interaction type corresponding to the second interaction action according to the second intermediate account number, the fourth intermediate account number, and the sixth intermediate account number; Determine the interaction type matching degree between the first interaction type and the second interaction type according to the type matching degree corresponding to the first interaction action and the type matching degree corresponding to the second interaction action.

9. A server system, characterized in that, Including a server, which is used to execute the method described in any one of claims 1-8.

Images