A sketch user behavior analysis method and system

Abstract

The present application provides a short play user behavior analysis method and system, relating to the technical field of data processing, the method comprising: step 1, obtaining the dynamic behavior data of the user on the short play platform through multi-dimensional data, including real-time watching behavior data and advertisement interaction data, and establishing a user behavior time sequence database; step 2, based on the user behavior time sequence database, extracting three fixed detection points, constructing a user behavior feature topology relationship with the three fixed detection points as the vertex, and analyzing the user behavior feature topology relationship to generate a behavior feature adjustment value; step 3, based on the behavior feature adjustment value, dynamically identifying the user's preference in entertainment, information, socialization and emotional needs, while constructing a psychological resistance evaluation mechanism to quantify the user's psychological resistance to advertisements, and dynamically updating the resistance level label. The present application understands the user's preference and watching habit, helps the short play platform to adjust the content creation, and improves the user retention rate and the platform business value.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for analyzing user behavior in short dramas. Background Technology

[0002] In the field of short drama user behavior analysis, traditional methods have limitations in dealing with complex user needs and dynamic scenarios. Relying on static tags makes it difficult to capture the real-time changes in user needs, leading to a disconnect between behavioral analysis and actual preferences. For example, a short drama platform primarily features "urban romance" and "suspense / mystery" content. Traditional methods tag users with "emotional preferences" based on their historical viewing time and continuously push similar content. However, in practice, user A recently experienced increased work pressure, shifting their viewing behavior from "romance dramas" to "stress-relieving comedy short dramas." Despite this, the traditional method, failing to update tags in real time, continued to push emotional content.

[0003] Furthermore, the lack of dynamic assessment of user psychological resistance makes it difficult for advertising strategies to balance user experience and commercial needs, leading to fluctuations in paid conversion rates. The platform inserts in-feed ads during short drama playback. Traditional methods deliver ads at a fixed frequency (e.g., once every 10 minutes), without considering users' real-time resistance to the ads. User C frequently skips ads during climaxes of the plot. Because the traditional method fails to identify the peak resistance during the "plot immersion period," it continues to deliver ads at a fixed frequency, resulting in insufficient ad click-through rates. Ultimately, the user disables the platform's ad push function. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method and system for analyzing user behavior in short dramas, so as to identify user needs and preferences and the degree of resistance to advertising, thereby optimizing content and advertising strategies, improving paid conversion rates and enhancing user stickiness.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a method for analyzing user behavior in short dramas, the method comprising:

[0007] Step 1: Obtain dynamic behavioral data of users on the short drama platform through multi-dimensional data, including real-time viewing behavior data and advertising interaction data, and establish a time-series database of user behavior.

[0008] Step 2: Based on the user behavior time series database, extract three fixed detection points, construct the user behavior feature topology relationship with the three fixed detection points as vertices, analyze the user behavior feature topology relationship, and generate behavior feature adjustment values;

[0009] Step 3: Based on behavioral characteristic adjustment values, dynamically identify users' preferences in entertainment, information, social and emotional needs, and at the same time build a psychological resistance assessment mechanism to quantify the degree of users' psychological resistance to advertising and dynamically update the resistance level label.

[0010] Step 4: Based on the resistance level tags, quantify the short drama's plot structure, performance artistry, and production technical parameters, and combine them with the paid conversion rate to construct a dynamic evaluation matrix and generate a content differentiation feature vector.

[0011] Step 5: Based on the content differentiation feature vector, analyze the dynamic relationship between users' social and payment behaviors through the identity recognition association mechanism, and adjust the advertising content matching strategy for users with strong psychological resistance to obtain advertising content matching results.

[0012] Step 6: Based on the ad content matching results, construct a dynamic prediction mechanism for user payment behavior and generate a paid recommendation strategy; at the same time, establish a closed-loop feedback channel for ad performance, and adjust the psychological resistance assessment parameters, content quality assessment indicators and payment prediction judgment logic by tracking the changing trend of ad avoidance in real time.

[0013] Furthermore, based on the user behavior time-series database, three fixed detection points are extracted. A topological relationship of user behavior features is constructed using these three fixed detection points as vertices. This topological relationship is then analyzed to generate adjusted behavioral feature values, including:

[0014] Extract the user's starting behavior point, advertising touch point, and ending behavior point during the short drama viewing process from the user behavior time series database, and use them as three fixed detection points;

[0015] Based on three fixed detection points, a user behavior connection relationship is constructed with the three fixed detection points as vertices according to their arrangement order in the user behavior time sequence.

[0016] By utilizing user behavior connections, the frequency of behavior transitions between vertices in the graph is analyzed to determine the strength of behavior associations, and the time interval distribution between adjacent vertices is calculated to extract time interval features.

[0017] By combining the strength of behavioral associations and time interval characteristics, weighted fusion calculations are used to generate behavioral feature adjustment values ​​that represent user behavior patterns.

[0018] Furthermore, based on behavioral characteristic adjustment values, user preferences in entertainment, information, social, and emotional needs are dynamically identified. Simultaneously, a psychological resistance assessment mechanism is constructed to quantify the degree of user psychological resistance to advertising and dynamically update resistance level labels, including:

[0019] Based on behavioral characteristic adjustment values, and through preset mapping rules, the strength of users' preference tendencies in the dimensions of entertainment needs, information needs, social needs, and emotional needs is identified and quantified.

[0020] Based on the strength of preference, key indicators are extracted from the ad interaction data, including ad skip rate and ad closing frequency, and the user's psychological resistance score is calculated through a predefined psychological resistance assessment algorithm.

[0021] Based on the user's psychological resistance score, according to the preset score range threshold, the user's psychological resistance is divided into multiple levels, and a corresponding initial resistance level label is assigned to the user.

[0022] Monitor new user behavior characteristics and advertising interaction data, repeatedly perform demand identification, psychological resistance calculation and level classification, and dynamically update the user's resistance level label based on the latest results.

[0023] Furthermore, based on resistance level tags, the plot structure, performance artistry, and production technology parameters of the short dramas are quantified, and a dynamic evaluation matrix is ​​constructed in conjunction with paid conversion rates to generate content differentiation feature vectors, including:

[0024] Based on user interruption behavior data in the resistance level tags, calculate the quantitative index of the short drama's plot coherence; based on user interaction behavior data, calculate the performance attractiveness score; and combine production specification data to generate production technical parameters.

[0025] The quantitative indicators of plot coherence, performance appeal scores, and production technical parameters are spatiotemporally aligned with the real-time paid conversion rate dataset.

[0026] The aligned multi-dimensional data is dynamically weighted and fused according to a preset time window to construct a dynamic evaluation matrix whose row and column dimensions respectively represent content attributes and time series.

[0027] Principal component decomposition is performed on the dynamic evaluation matrix to determine the main feature combinations with relatively prominent eigenvalues. The extracted feature combinations are then normalized to generate feature vectors that characterize the differences in the content of short dramas.

[0028] Furthermore, based on content-differentiated feature vectors, the dynamic relationship between users' social interactions and payment behaviors is analyzed through identity recognition association mechanisms. The advertising content matching strategy is adjusted for users with strong psychological resistance to obtain advertising content matching results, including:

[0029] By utilizing the identity association mechanism, we analyze the correlation between differentiated content feature vectors and users' social behavior data, and obtain the first analysis results representing the correlation between user identity and social behavior.

[0030] Based on the first analysis results, predict the potential payment behavior patterns of users and generate payment behavior prediction results, including predictions of payment behavior tendencies.

[0031] Based on the resistance level labels, identify and determine the target users marked as having prominent psychological resistance from the user group, and generate a target user identifier set;

[0032] Based on the paid behavior prediction results, an advertising strategy is generated for the target users identified by the target user identifier set.

[0033] Based on the adjusted advertising strategy, the final ad content matching results are generated, including personalized ad content types and delivery timing configurations for different users.

[0034] Furthermore, based on the ad content matching results, a dynamic prediction mechanism for user payment behavior is constructed to generate a paid recommendation strategy. Simultaneously, a closed-loop feedback channel for ad performance is established. By tracking the changing trends of ad avoidance in real time, the psychological resistance assessment parameters, content quality assessment indicators, and the judgment logic for paid prediction are adjusted, including:

[0035] Based on the ad content matching results, user interaction data with ads is extracted to form a user behavior dataset.

[0036] By utilizing user behavior datasets, a dynamic rule system for predicting user payment behavior is established to obtain payment behavior prediction rules;

[0037] Based on paid behavior prediction rules, the paid behavior of target users is predicted, and a paid recommendation strategy is generated.

[0038] Based on the paid recommendation strategy, a closed-loop feedback channel for ad performance is established to collect data on user avoidance behavior towards ads in real time, and based on the avoidance behavior data, the trend data of ad avoidance rate is calculated.

[0039] Based on trend data, we adaptively adjust the parameter values ​​in psychological resistance assessment, the indicator items in content quality assessment, and the judgment logic parameters of the paid behavior prediction rules.

[0040] Furthermore, the paid recommendation strategy includes the type of recommended ad content, the timing of recommendation delivery, and the recommendation intensity.

[0041] Secondly, a short drama user behavior analysis system includes:

[0042] The data acquisition module is used to collect dynamic behavioral data of users on the short drama platform, including real-time viewing behavior data and advertising interaction data, and to establish a time-series database of user behavior.

[0043] The behavior construction module is used to extract three fixed detection points based on the user behavior time series database, construct the user behavior feature topology relationship with the three fixed detection points as vertices, analyze the user behavior feature topology relationship, and generate behavior feature adjustment values.

[0044] The resistance assessment module is used to dynamically identify users' preferences in entertainment, information, social and emotional needs based on behavioral characteristic adjustment values. At the same time, it constructs a psychological resistance assessment mechanism to quantify the degree of users' psychological resistance to advertising and dynamically update the resistance level label.

[0045] The differentiated evaluation module is used to quantify the plot structure, performance artistry, and production technical parameters of short dramas based on resistance level tags, and to build a dynamic evaluation matrix by combining paid conversion rate to generate content differentiation feature vectors.

[0046] The matching strategy module is used to analyze the dynamic relationship between users' social and payment behaviors based on differentiated feature vectors and identity recognition association mechanisms, and to adjust the advertising content matching strategy for users with strong psychological resistance in order to obtain advertising content matching results.

[0047] The paid prediction module is used to build a dynamic prediction mechanism for user payment behavior based on the ad content matching results and generate paid recommendation strategies. At the same time, it establishes a closed-loop feedback channel for ad performance, and adjusts the psychological resistance assessment parameters, content quality assessment indicators and paid prediction judgment logic by tracking the changing trend of ad avoidance in real time.

[0048] Thirdly, a computing device includes:

[0049] One or more processors;

[0050] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0051] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0052] The above-described solution of the present invention has at least the following beneficial effects:

[0053] By collecting real-time user viewing behavior data and advertising interaction data, a time-series database is established to comprehensively cover the dynamic behavioral trajectory of users on the short drama platform. Compared with traditional single-dimensional user behavior analysis, this method constructs a topological relationship of behavioral features from multiple key nodes such as start, ad contact, and termination. Combined with behavioral correlation strength and time interval features, adjustment values ​​are generated, which can more accurately capture user behavior patterns. Based on the behavioral feature adjustment values, the platform dynamically identifies users' needs and preferences in entertainment, information, social interaction, and emotion, and constructs a psychological resistance assessment mechanism to quantify the degree of user resistance to advertising. This allows the platform to deeply understand users' inner needs and provide personalized content recommendations and advertising for users with different needs and resistance levels, thereby improving user experience and platform service quality.

[0054] By combining user resistance level tags, quantifying the plot structure, performance artistry, and production technical parameters of short dramas, and integrating paid conversion rates to construct a dynamic evaluation matrix, this approach can scientifically assess the content quality and commercial value of short dramas. It provides content creators with clear creative direction guidance, while helping operators optimize resource allocation, improve the efficiency of high-quality content production, and enhance the platform's content competitiveness. Based on user behavior and content differentiation characteristics, it analyzes the dynamic connection between user social interaction and paid behavior through an identity recognition association mechanism. For users with strong psychological resistance, it adjusts advertising content matching strategies. Compared to traditional advertising methods, this method effectively reduces user resistance to advertising, improves the alignment between advertising and user needs, thereby increasing ad click-through rates and conversion rates, bringing higher ROI to advertisers, and increasing platform advertising revenue. Furthermore, it establishes a dynamic prediction mechanism for user paid behavior and a closed-loop feedback channel for ad performance, tracking ad avoidance in real time and dynamically adjusting psychological resistance assessment parameters, content quality assessment indicators, and paid prediction judgment logic. This allows the platform to adapt to market changes and evolving user needs in a timely manner, maintaining a competitive advantage. Attached Figure Description

[0055] Figure 1 This is a flowchart illustrating a short drama user behavior analysis method provided by an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of a short drama user behavior analysis system provided by an embodiment of the present invention. Detailed Implementation

[0057] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0058] like Figure 1 As shown, an embodiment of the present invention proposes a method for analyzing user behavior in short dramas, the method comprising the following steps:

[0059] Step 1: Obtain dynamic behavioral data of users on the short drama platform through multi-dimensional data, including real-time viewing behavior data and advertising interaction data, and establish a time-series database of user behavior.

[0060] Step 2: Based on the user behavior time series database, extract three fixed detection points, construct the user behavior feature topology relationship with the three fixed detection points as vertices, analyze the user behavior feature topology relationship, and generate behavior feature adjustment values;

[0061] Step 3: Based on behavioral characteristic adjustment values, dynamically identify users' preferences in entertainment, information, social and emotional needs, and at the same time build a psychological resistance assessment mechanism to quantify the degree of users' psychological resistance to advertising and dynamically update the resistance level label.

[0062] Step 4: Based on the resistance level tags, quantify the short drama's plot structure, performance artistry, and production technical parameters, and combine them with the paid conversion rate to construct a dynamic evaluation matrix and generate a content differentiation feature vector.

[0063] Step 5: Based on the content differentiation feature vector, analyze the dynamic relationship between users' social and payment behaviors through the identity recognition association mechanism, and adjust the advertising content matching strategy for users with strong psychological resistance to obtain advertising content matching results.

[0064] Step 6: Based on the ad content matching results, construct a dynamic prediction mechanism for user payment behavior and generate a paid recommendation strategy; at the same time, establish a closed-loop feedback channel for ad performance, and adjust the psychological resistance assessment parameters, content quality assessment indicators and payment prediction judgment logic by tracking the changing trend of ad avoidance in real time.

[0065] In this embodiment of the invention, by collecting real-time viewing behavior data and advertising interaction data from multiple dimensions and establishing a time-series database, comprehensive, real-time, and dynamic storage of user behavior data is achieved. Compared with traditional single data collection methods, this step can completely record the user's entire behavioral trajectory from the start to the end of viewing, avoiding the omission of key behavioral information, improving the integrity and usability of data, and laying a solid foundation for accurate analysis. Based on the time-series database, three fixed detection points are extracted to construct behavioral feature topological relationships, breaking through the traditional linear analysis mode. The correlation between user behaviors is presented from a graphical and structured perspective. By analyzing the frequency and time interval characteristics of behavioral transitions between vertices, the regularity and differences of user behavior patterns can be quantified. The generated behavioral feature adjustment values ​​can intuitively reflect the characteristics of user behavior, effectively improving the scientificity and accuracy of user behavior analysis. Based on the behavioral feature adjustment values, multi-dimensional user needs and preferences are dynamically identified, and the degree of psychological resistance to advertising is quantified. This achieves in-depth mining from behavioral data to the user's inner needs and emotional attitudes. Dynamically updating the resistance level label can track changes in the user's attitude towards advertising in real time, changing the limitations of traditional static analysis. This enables the platform to accurately grasp the dynamics of user needs and advertising acceptance, enhancing user experience and the targeting of platform services.

[0066] By combining resistance level tags to quantify multiple parameters of short dramas and constructing a dynamic evaluation matrix, a deep integration of content quality assessment, user feedback, and commercial value is achieved. Through principal component decomposition, content-differentiated feature vectors are generated, enabling the scientific extraction of the core strengths and weaknesses of short drama content. This provides content creators with data-driven optimization directions, while simultaneously helping platform operators accurately assess the commercial potential of content, optimize resource allocation, and enhance the overall competitiveness and market adaptability of the platform's content. Based on the analysis of content-differentiated feature vectors, the dynamic relationship between user social and payment behaviors is analyzed, and advertising matching strategies are adjusted for high-resistance users, achieving a shift in advertising from "broadcasting" to "precise targeting." This is achieved through an identity-based association mechanism. It can uncover users' potential payment motives and social media value, customize personalized advertising content and timing for different users, effectively reduce user resistance to advertising, improve the matching degree between advertising and user needs, and thus increase advertising conversion rate and user willingness to pay, achieving a win-win situation for the platform, advertisers and users. It builds a dynamic prediction mechanism for user payment behavior and a closed-loop feedback channel for advertising performance, realizing real-time dynamic optimization of operational strategies. By tracking advertising avoidance trends and adjusting evaluation parameters and judgment logic, it can quickly respond to market changes and the evolution of user needs, avoiding user churn and loss of business opportunities due to strategy lag. The generated paid recommendation strategy, combined with advertising matching results, can accurately guide user payment behavior.

[0067] In a preferred embodiment of the present invention, step 1 above, which involves acquiring dynamic behavioral data of users on the short drama platform through multi-dimensional data, including real-time viewing behavior data and advertising interaction data, and establishing a user behavior time-series database, may include:

[0068] In this embodiment of the invention, real-time viewing behavior data mainly comes from the log records of the playback module, including user actions such as clicking play, pause, fast forward, rewind, dragging the progress bar, viewing duration, number of episodes watched, and switching episodes. Advertising interaction data is recorded by the advertising delivery system, covering behaviors such as ad display, click, skip, close, and viewing duration. The data collection program is triggered every time a user performs a behavior. For example, when user A clicks to play the short drama "XXXX", the playback module immediately generates a record containing user ID (e.g., UID12345), operation type ("play"), short drama ID (e.g., DRAMA001), operation time (accurate to the second, e.g., December 15, 2024, 14:23:10), device type (e.g., "Android phone"), device model (e.g., "Xiaomi 13"). If the user pauses the viewing, a new data record is recorded again, containing the same user ID and short drama ID, but the operation type is changed to "pause", and a new operation time (e.g., December 15, 2024, 14:25:30).

[0069] Regarding advertising interaction data, when the platform displays a beauty advertisement to user B, the advertising delivery system records information such as user ID (UID67890), advertisement ID (AD0001), operation type ("display"), and display time (December 15, 2024, 14:30:00). If user B clicks "skip" 5 seconds after the advertisement plays, a new record will be added, with the operation type changed to "skip," the recording time (December 15, 2024, 14:30:05), and the advertisement ID and user ID corresponding to the skip operation. The collected data will first enter a temporary storage area for preliminary data cleaning. The cleaning process includes checking the integrity of the data and removing records with missing key information (such as no user ID or operation time); identifying and deleting duplicate records, such as data repeatedly reported for the same operation due to network latency; and processing abnormal data, such as correcting or marking obviously illogical viewing durations (such as negative durations or durations exceeding several times the total duration of the short drama) as invalid data.

[0070] After data cleaning, the data will be imported into a user behavior time-series database. This database uses a relational database design and contains at least two core tables: a user behavior record table and a metadata table. The user behavior record table stores specific behavioral data, and its structure includes fields such as record ID (uniquely identifying each record), user ID, operation type, short drama ID, advertisement ID (if advertising interaction is involved), operation time, device type, and device model. The metadata table records basic user information, such as user registration time, gender, age, and region. During data storage, user behavior records are sorted chronologically. For example, user C, while watching a short drama, successively played, paused, and fast-forwarded. Operations such as continuing playback are recorded and stored sequentially in the database according to the order of their occurrence, ensuring a complete behavioral timeline. To improve data query and analysis efficiency, indexes are created for the user ID and operation time fields. As users continuously generate new behavioral data, this data is inserted into the database in real time. During insertion, data validation is performed again to ensure the new data conforms to the database table structure requirements. For example, the operation time of a new record must be later than the time of the user's previous record; otherwise, it is considered abnormal data and intercepted and processed. In this way, the user behavior timeline database can be continuously updated and improved, continuously accumulating dynamic behavioral data of users on the short drama platform.

[0071] In a preferred embodiment of the present invention, step 2 above, which involves extracting three fixed detection points based on a user behavior time-series database, constructing a user behavior feature topology using these three fixed detection points as vertices, and analyzing the user behavior feature topology to generate behavior feature adjustment values, may include:

[0072] Step 220: Extract the user's starting behavior point, advertising touch point and ending behavior point during the short drama viewing process from the user behavior time series database, and use them as three fixed detection points;

[0073] Step 221: Based on the three fixed detection points, construct the user behavior connection relationship with the three fixed detection points as vertices according to their arrangement order in the user behavior time sequence;

[0074] Step 222: Utilize user behavior connection relationships to analyze the frequency of behavior transitions between vertices in the graph to determine the strength of behavior associations, and calculate the time interval distribution between adjacent vertices to extract time interval features;

[0075] Step 223: Combining the strength of behavioral association and time interval features, a weighted fusion calculation is performed to generate behavioral feature adjustment values ​​that represent user behavior patterns.

[0076] In this embodiment of the invention, each piece of user behavior data in the user behavior time-series database carries detailed attribute information. In addition to the timestamp accurate to the second, it also includes behavior type (such as "playing a short drama", "clicking an advertisement", "pausing to watch"), operation object (specific short drama name, advertisement number), device information (phone model, system version), etc. When the search begins, all behavior records related to watching short dramas by a certain user within a specific time period (such as the last 30 days) are first filtered out. Taking the search for the starting behavior point as an example, these records are checked one by one in order of the timestamp from smallest to largest. When the first behavior record of the "playing a short drama" type is found, the timestamp, behavior type, short drama name and other information corresponding to the record are extracted and determined as the starting behavior point. For example, if user A clicks to play the short drama "XXXX" for the first time at 15:23:10 on October 1, 2024, then this time point and related behavior information constitute user A's starting behavior point.

[0077] For extracting ad touchpoints, the system specifically searches for records containing behavior types such as "ad display," "ad click," and "skip ad" from the filtered behavior records. Once a record showing the user's first interaction with an ad is found—for example, if user A clicks on an ad for the first time at 15:28:30 after watching "XXXX" for 5 minutes—the timestamp and behavior details of that record are extracted and set as the ad touchpoint. Determining the termination behavior point involves finding the user's last behavior record related to watching the short drama within that time period. When it is detected that the user has not performed any actions related to the short drama for a certain period (e.g., 10 minutes), or exhibits behaviors that clearly indicate the end of viewing, such as "exiting the application" or "closing the playback page," the time point of that record is identified as the termination behavior point. For example, if user A closes the playback page at 15:45:20, this time point becomes their termination behavior point.

[0078] After obtaining three fixed detection points, a data structure is created to store and manage the relationships between these points. Taking user A as an example, these three points are added to the data structure as nodes according to the time sequence of the starting behavior point (15:23:10), the ad touchpoint (15:28:30), and the ending behavior point (15:45:20). Then, connections are established between these nodes, with the direction of the connections following the time sequence from the starting behavior point to the ad touchpoint, and then from the ad touchpoint to the ending behavior point. To make the display more intuitive, this relationship can be visualized by connecting the three nodes with arrow segments on the interface. The direction of the arrows clearly indicates the order of the behaviors. At the same time, the data structure records the detailed information of each node and the attributes of the connections between the nodes (such as connection time, behavior type, etc.), forming a complete layout of user behavior connection relationships.

[0079] To calculate the frequency of behavioral transitions between vertices, a statistical period is first defined, such as a week. Within this period, the behavioral connection data of all users in the database is traversed. Taking the calculation of the transition frequency from the starting behavior point to the ad touchpoint as an example, the number of users who reach the ad touchpoint within a certain time range (e.g., within 10 minutes after watching a short drama) from the starting behavior point is counted. Suppose that in a certain week, a total of 10,000 users watch short dramas, and 3,000 of them interact with the ad within 10 minutes after watching the short drama (reaching the ad touchpoint). Then, the behavioral transition frequency from the starting behavior point to the ad touchpoint is 3,000 ÷ 10,000 × 100% = 30%. For the transition frequency between other vertices, such as from the ad touchpoint to the ending behavior point, a similar statistical method is used. By calculating the proportion of users who transition to the total number of users, the corresponding behavioral transition frequency is obtained, thereby measuring the degree of correlation between different behavioral stages.

[0080] When calculating the time interval between adjacent vertices, for each user's behavioral connection, the time difference from the starting behavior point to the ad touchpoint and from the ad touchpoint to the ending behavior point are calculated separately. For example, if user A's starting behavior point is 15:23:10 and the ad touchpoint is 15:28:30, then the time interval between them is 5 minutes and 20 seconds (320 seconds). These time interval data from all users are collected and grouped for statistical analysis. Assuming the time intervals are divided into several intervals such as 0-300 seconds, 301-600 seconds, and 601-900 seconds, the number of time interval data contained in each interval is counted. For example, the statistics show that among all users, 4000 time interval data fall within the 0-300 second interval, and 3000 fall within the 301-600 second interval. In this way, the distribution of time intervals between adjacent vertices can be clearly presented, and the time consumption pattern of users in different behavioral stages can be understood.

[0081] Before generating the behavioral feature adjustment value, the platform operations team determines the weights of behavioral association strength and time interval features based on business needs and data analysis. For example, after analysis, it was found that users' behavioral conversion tendencies have a more critical impact on ad placement and content recommendation. Therefore, the weight of behavioral association strength is set to 0.7, and the weight of time interval features is set to 0.3. The previously calculated behavioral conversion frequency and time interval distribution data are standardized to make them fall within the same numerical range (such as 0-1). Taking user A as an example, assuming that the standardized behavioral conversion frequency from the starting behavior point to the ad touchpoint is 0.8, and the standardized time interval feature from the ad touchpoint to the ending behavior point is 0.6, then, the weighted calculation is performed according to the set weights. For user A, the behavioral feature adjustment value = behavioral association strength (0.8) × weight (0.7) + time interval feature (0.6) × weight (0.3) = 0.8 × 0.7 + 0.6 × 0.3 = 0.56 + 0.18 = 0.74. This weighted fusion method generates a behavioral feature adjustment value for each user that comprehensively reflects their behavioral patterns. This value can be used for user behavior analysis and classification.

[0082] By accurately extracting the starting behavior point, advertising touchpoint, and ending behavior point, the complete cycle of user viewing behavior in short dramas can be clearly defined, avoiding analytical biases caused by incomplete data extraction. This makes user behavior research more targeted and systematic, laying the foundation for a comprehensive understanding of user behavior trajectories during short drama viewing. Constructing user behavior connections overcomes the limitations of traditional data list analysis, quickly identifying common patterns and abnormal paths in user behavior, and helping to discover potential patterns in user behavior. Analyzing behavior transition frequency and time interval characteristics allows for the quantification of the intrinsic connections and temporal patterns between user behaviors. Calculating behavior transition frequency helps the platform understand user tendencies at different behavioral stages, such as determining whether users are more inclined to… Immediately stopping viewing when an ad appears allows for targeted optimization of ad delivery strategies; analysis of time interval characteristics reveals users' time consumption habits during short drama viewing, such as the preferred viewing time, providing data support for content creation duration design and pacing control, effectively improving the scientific nature and accuracy of platform operation strategies; generating behavioral characteristic adjustment values, which weight and fuse behavioral correlation strength and time interval characteristics, achieves the quantification and comprehensive expression of user behavioral characteristics. This single value can concisely and comprehensively summarize the complex behavioral patterns of users. By analyzing different user behavioral characteristic adjustment values, the platform can quickly distinguish behavioral differences among user groups, improving user experience and platform operational efficiency.

[0083] In a preferred embodiment of the present invention, step 3 above, which dynamically identifies user preferences in entertainment, information, social, and emotional needs based on behavioral characteristic adjustment values, and simultaneously constructs a psychological resistance assessment mechanism to quantify the degree of user psychological resistance to advertising and dynamically updates the resistance level label, may include:

[0084] Step 330: Based on behavioral characteristic adjustment values, identify and quantify the intensity of user preference tendencies in the dimensions of entertainment needs, information needs, social needs, and emotional needs through preset mapping rules;

[0085] Step 331: Based on the strength of preference tendency, extract key indicators from the advertising interaction data, including ad skip rate and ad closing frequency, and calculate the user's psychological resistance score through a predefined psychological resistance assessment algorithm;

[0086] Step 332: Based on the user's psychological resistance score, according to the preset score range threshold, the user's psychological resistance is divided into multiple levels, and a corresponding initial resistance level label is assigned to the user.

[0087] Step 333: Monitor new user behavior characteristics and advertising interaction data, repeatedly perform demand identification, psychological resistance calculation and level classification, and dynamically update the user's resistance level label based on the latest results.

[0088] In this embodiment of the invention, when identifying user needs and preferences using behavioral feature adjustment values, a search mechanism based on the Sparrow Search algorithm is introduced. Each user is considered a "sparrow," and the user's behavioral feature adjustment value serves as the sparrow's initial position in the search space. The platform's preset four needs dimensions (entertainment, information, social, and emotional) constitute different areas of the search space, while the preset mapping rules are equivalent to "environmental cues" guiding the sparrow's flight. When the algorithm starts, all "sparrows" (users) locate themselves in the search space based on their own behavioral feature adjustment values, much like a flock of sparrows foraging for food. A portion of "explorer" users (randomly selected at a certain percentage, such as 20%) first explore between the areas of each needs dimension according to the mapping rules. For example, if explorer user A's behavioral feature adjustment value is 0.6, it will prioritize exploring the "high preference" area of ​​the entertainment needs dimension and the "medium-high preference" area of ​​the emotional needs dimension, evaluating the strength of the preference tendency corresponding to that position.

[0089] The remaining "follower" users (80%) observe the explorer's results. If explorer user B achieves a high "fitness" in the "high preference" area of ​​the information demand dimension (i.e., the area matches the user's behavior pattern well), the follower users will adjust their position in the search space according to the following strategy of the Sparrow Search algorithm, move closer to that area, and re-evaluate the strength of their preference. Each user can find the strength of their preference in the demand dimension that best matches their own behavior pattern, thus completing the quantification process.

[0090] When calculating the score of psychological resistance, the ad skip rate and ad closing frequency are considered as key factors affecting the "sparrow's survival." A competitive environment is constructed by combining the strength of user preference tendencies. The strength of each user's preference tendencies will affect their "base energy value" in the competition. For example, users with a "high preference" for entertainment needs have a higher initial energy value when facing entertainment ads, and are theoretically more receptive to the ads. After extracting ad interaction data, the ad skip rate and ad closing frequency of each user are calculated. Just as sparrows face the threat of predators when foraging for food, ad skipping and closing behaviors will consume the user's "energy value." The predefined psychological resistance assessment algorithm then simulates the sparrow's competitive rules: skipping an ad once will cause the user's energy value in the corresponding need dimension to decrease by a certain percentage (e.g., 15%); closing an ad once will cause the energy value to decrease by an even higher percentage (e.g., 25%).

[0091] In this competitive environment, all users influence each other. Some "weak sparrows" (users with high resistance) with lower energy values ​​will trigger anti-predation strategies in the sparrow search algorithm, such as actively changing their position in the search space (adjusting their attitude or behavior patterns towards ads). After multiple rounds of competition, the user's final remaining energy value is converted into a psychological resistance score. For example, user C's initial energy value is 80, and after being consumed by ad interaction, 30 remains. This resistance score is calculated according to a preset conversion rule (e.g., score = ...). The psychological resistance score was 0.375.

[0092] The platform's preset score range thresholds act like different "habitats" to accommodate users, like sparrows, with varying levels of resistance. Users are assigned to corresponding habitats (resistance levels) based on their psychological resistance scores. For example, users with scores between 0 and 0.3 enter the "low resistance" habitat, while those between 0.3 and 0.6 enter the "medium-low resistance" habitat. When assigning initial resistance level labels, the platform borrows from the group classification habits of sparrows, creating a "sparrow flock" among users in the same habitat. Users within each flock are assigned the same resistance level label. Simultaneously, the platform records the characteristics of each flock, such as the "high resistance" flock, which typically exhibits high ad skip rates and low preference intensity. These characteristics are used for strategy development. The platform continuously monitors new user behavior and ad interaction data, much like observing sparrow migration behavior in different seasons. When a user exhibits new behavior, their position in the search space (behavioral characteristic adjustment value) and energy value (psychological resistance level) change.

[0093] If a user's new behavior causes their psychological resistance score to exceed the range of their original habitat (resistance level), the user, like a migrating sparrow, will leave the original flock and enter a new habitat. For example, user D originally belonged to the "low to medium resistance" group. Due to frequently skipping ads, their psychological resistance score rose to 0.7, and they were migrated to the "high to medium resistance" group, with their resistance level label updated. During the migration process, the user will adjust their behavior based on the environment of the new habitat (the behavioral patterns of the new resistance level group), referring to the learning mechanism of the sparrow search algorithm. At the same time, based on the dynamic changes of the entire "sparrow population" (user group), the user will adjust the preset mapping rules, psychological resistance assessment algorithm parameters, etc., forming an adaptive dynamic update closed loop.

[0094] By integrating the Sparrow Search algorithm into user demand identification and psychological resistance assessment, this approach breaks through the static model of traditional data processing. Utilizing a bio-inspired dynamic search and competition mechanism, it achieves efficient analysis of user behavior data. Compared to traditional methods, this fusion approach is akin to giving the system "intelligent navigation," quickly locating user needs and preferences as well as psychological resistance states, significantly shortening analysis time and improving data processing efficiency, making it particularly suitable for large-scale user data scenarios. The Sparrow Search algorithm's collaborative and dynamic adjustment characteristics can accurately capture subtle changes in user behavior. Whether it's the explorer's initial probing or the follower's learning and adjustment, it ensures a more accurate quantification of user needs and preferences and a more precise calculation of psychological resistance. In practice, the dynamic migration and tag update mechanism based on changes in user behavior has stronger environmental adaptability, enabling it to respond promptly to market changes and evolving user needs, and maintain the accuracy of analysis results. By simulating sparrow classification and competition strategies, the platform can more accurately segment user groups and develop differentiated advertising strategies for users with different levels of resistance. For example, it can reduce advertising interference for "high-resistance" users and push content that better meets their needs; and it can reasonably increase advertising exposure for "low-resistance" users to improve advertising conversion rates. This precise operation model can not only improve user experience and enhance user stickiness, but also create higher commercial value for the platform, achieving a win-win situation for both users and the platform.

[0095] In a preferred embodiment of the present invention, step 4 above, which quantifies the plot structure, performance artistry, and production technical parameters of the short drama based on resistance level tags, and constructs a dynamic evaluation matrix in conjunction with paid conversion rates to generate content differentiation feature vectors, may include:

[0096] Step 440: Calculate the plot coherence quantification index of the short drama based on the user interruption behavior data in the resistance level label; calculate the performance attractiveness score based on the user interaction behavior data; and generate production technical parameters by combining the production specification data.

[0097] Step 441: Spatiotemporally align the quantitative indicators of plot coherence, performance appeal scores, and production technical parameters with the real-time acquired paid conversion rate dataset.

[0098] Step 442: The aligned multi-dimensional data is dynamically weighted and fused according to a preset time window to construct a dynamic evaluation matrix in which the row and column dimensions respectively represent the content attributes and the time series.

[0099] Step 443: Perform principal component decomposition on the dynamic evaluation matrix to determine the main feature combinations with relatively prominent eigenvalues, and normalize the extracted feature combinations to generate feature vectors that characterize the differences in the content of the short drama.

[0100] In this embodiment of the invention, user interruption behavior data with resistance level labels are first filtered from the user behavior time-series database. Interruption behaviors include user-initiated pausing, exiting viewing, and fast-forwarding for a certain duration (e.g., exceeding 10% of the current playback progress). Taking a short drama series "XXXX" as an example, interruption behavior records of all users watching the series within a week are statistically analyzed. For each user, if an interruption occurs during viewing, the interruption time and the resumption time are recorded (if no resumption occurs, it is considered a complete interruption). By calculating the average duration between two adjacent interruptions, the user's perceived continuity of the plot is preliminarily assessed. For example, if user A watched "XXXX" and experienced three interruptions with intervals of 10 minutes, 15 minutes, and 8 minutes respectively, then their average interruption interval would be (10 + 15 + 8). ÷3=11 minutes; Next, the average interruption interval of all users is summarized, and the overall average and standard deviation are calculated. If the overall average interruption interval of the short drama is relatively long and the standard deviation is small, it means that most users watch it relatively smoothly and the plot is coherent. Conversely, if the average interval is short and the standard deviation is large, it indicates that the plot is not coherent. Assuming that the average interruption interval of all users of "XXXX" is 8 minutes and the standard deviation is 3 minutes, according to the preset scoring rules (such as average interval >10 minutes is excellent, 8-10 minutes is good, 6-8 minutes is average, and <6 minutes is poor), its plot coherence is quantified as "good" and assigned a corresponding score, such as 70 points.

[0101] Extract user interaction data, including likes, comments, shares, and favorites. Using "XXXX" as an example, count the total number of likes, comments, shares, and favorites received by the short drama within a week. To reflect the different levels of importance of each interaction, assign different weights to each behavior: likes 0.2, comments 0.3, shares 0.4, and favorites 0.1. Assuming "XXXX" received 1000 likes, 200 comments, 300 shares, and 150 favorites within a week, first standardize the quantity of each behavior; for example, the standardized value of likes = Assuming the standardized value is 1.2; Standardized comment value = =1.5; Shared normalized value= =1.3; Standardized value for collections = =1.1; Then, calculate the performance appeal score according to the weights: Score = Normalized value of likes × 0.2 + Normalized value of comments × 0.3 + Normalized value of shares × 0.4 + Normalized value of favorites × 0.1 = 1.2 × 0.2 + 1.5 × 0.3 + 1.3 × 0.4 + 1.1 × 0.1 = 0.24 + 0.45 + 0.52 + 0.11 = 1.32. Then map this score to a 0-100 point scale. For example, after mapping using the formula, the score = 1.32 × 75 (assuming the mapping coefficient is 75, it can be adjusted according to the actual situation) = 99 points, that is, the performance appeal score of "XXXX" is 99 points.

[0102] The platform retrieves short drama production specifications from its content management database, including video resolution (e.g., 1080P, 4K), frame rate (e.g., 24 fps, 60 fps), audio sampling rate (e.g., 44.1kHz, 48kHz), and the number of special effects used (e.g., the percentage of special effects shots). For each specification, based on industry standards and user feedback, different parameter values ​​are assigned corresponding technical scores. For example, 4K resolution corresponds to 90 points, 1080P to 70 points; 60 fps to 85 points, 24 fps to 75 points. These technical scores are then distributed according to... The total score for production technical parameters is calculated by weighting the parameters according to certain weights (such as resolution weight 0.3, frame rate weight 0.2, audio sampling rate weight 0.2, and number of special effects used weight 0.3). For example, if a short drama has a resolution of 1080P, a frame rate of 60 frames / second, an audio sampling rate of 48kHz, and a special effects shot ratio of 20%, the corresponding technical scores are 70, 85, 80, and 70 points, respectively. Then the total score for production technical parameters is 70×0.3+85×0.2+80×0.2+70×0.3=21+17+16+21=75 points.

[0103] Real-time acquisition of paid conversion rate dataset for short dramas, paid conversion rate = For example, if the number of viewers of "XXXX" at a certain moment is 10,000, the number of paying users is 500, and the paid conversion rate is 5%, the quantitative indicators of plot coherence, performance attractiveness scores, and production technical parameters are all calculated within a certain time period (such as one week), while the paid conversion rate changes in real time. In order to effectively integrate these data, the data is aligned with the paid conversion rate at the most recent time point, with time as the axis. In specific operation, a time tolerance is set (such as 1 hour). If the timestamp of the paid conversion rate data differs from the end time of the calculation period of other indicators by less than 1 hour, it is directly matched; if it exceeds 1 hour, the closest paid conversion rate data within the tolerance range is identified and aligned. For example, if the calculation period of indicators such as plot coherence of a short drama ends at 23:00 on November 1, 2024, and the paid conversion rate data is the latest and within the tolerance range at 00:30 on November 2, 2024, then the paid conversion rate at that time point (let's say 6%) is aligned with other indicators of the short drama.

[0104] A preset time window, such as 3 days, is used. Within each time window, four data points—aligned plot coherence quantification, performance appeal score, production technical parameters, and paid conversion rate—are dynamically weighted and integrated according to certain weights. The weighting is adjusted based on the platform's operational strategy and data analysis. For example, plot coherence might have a weight of 0.2, performance appeal 0.3, production technical parameters 0.2, and paid conversion rate 0.3. Taking "XXXX" as an example, in the first 3-day time window, its plot coherence quantification score would be 7. With a score of 0, a performance appeal score of 99, a production technical parameter score of 75, and a paid conversion rate of 6% (equivalent to 6 points on a 100-point scale), the weighted fusion value is: 70×0.2+99×0.3+75×0.2+6×0.3=14+29.7+15+1.8=60.5 points. The weighted fusion value of each short drama within each time window is recorded to form a matrix. The rows of the matrix represent different short dramas, and the columns represent different time windows. Each element in the matrix is ​​the weighted fusion value of the corresponding short drama within the corresponding time window. This constructs a dynamic evaluation matrix.

[0105] When performing principal component decomposition on the dynamic evaluation matrix, the covariance matrix is ​​first calculated to measure the correlation between indicators. For example, if the covariance values ​​of plot coherence and performance appeal are high, it indicates a strong correlation between these two indicators. After calculating eigenvalues ​​and eigenvectors, the contribution rate of each eigenvalue is calculated (eigenvalue ÷ sum of all eigenvalues). Assuming the calculated three eigenvalues ​​are 1.8, 1.2, and 0.5, and the total eigenvalue is 3.5, their contribution rates are 1.8 ÷ 3.5 ≈ 0.514, 1.2 ÷ 3.5 ≈ 0.343, and 0.5 ÷ 3.5 ≈ 0.143, respectively. The two principal components with a cumulative contribution rate exceeding 85% are identified (i.e., the principal components corresponding to eigenvalues ​​1.8 and 1.2). When normalizing the extracted eigenvectors, the system uses the "max-min normalization method." Assuming the first eigenvector is [0.6, 0.4, 0.2, 0.1], The maximum value is 0.6 and the minimum value is 0.1. The normalized vector is [(0.6 - 0.1)÷(0.6 - 0.1), (0.4 - 0.1)÷(0.6 - 0.1), (0.2 - 0.1)÷(0.6 - 0.1), (0.1 - 0.1)÷(0.6 - 0.1)]=[1, 0.6, 0.2, 0]. The final normalized feature vector is the feature vector that represents the content differences of the short drama "XXXX".

[0106] By combining user resistance level tags to calculate short drama evaluation indicators, precise quantification of content quality is achieved by inferring from user behavior. Traditional content evaluation often relies on subjective evaluations or single data indicators. This step, however, utilizes multi-dimensional data such as user interruption behavior and interaction behavior to scientifically quantify plot coherence, performance appeal, and production technology level from the perspective of real user experience. For example, interruption behavior data can identify obvious plot breaks in a short drama, allowing for timely feedback to the creators for optimization and improving the relevance and quality of content creation. Data spatiotemporal alignment ensures the timeliness and consistency of multi-dimensional data, avoiding analytical biases caused by time misalignment. In the rapidly changing short drama market, combining real-time paid conversion rate data with content evaluation indicators can accurately reflect the correlation between content and user willingness to pay. For example, after adjusting the plot of a short drama, data alignment revealed an increase in paid conversion rate. The construction of a dynamic evaluation matrix will... By integrating and weighting dimensional data, a systematic content evaluation system is formed. The dynamic processing method based on time windows can capture the changing trends of short drama content performance. For example, it can observe the rating fluctuations of a short drama over different time periods and analyze the evolution of its popularity. Simultaneously, the weighted fusion allocates weights according to the importance of each indicator, highlighting the impact of key factors on content value, making the evaluation results more aligned with actual business needs. It generates content-differentiated feature vectors, simplifying complex multi-dimensional data into comparable feature representations, facilitating the platform's rapid classification and screening of a large number of short dramas. Principal component decomposition extracts key features, removing redundant information and retaining the elements that best reflect the characteristics of the short drama. For example, the platform can quickly identify short dramas with unique styles based on feature vectors for accurate recommendations, improving the efficiency of users discovering high-quality content and enhancing the platform's content competitiveness. It also provides clear content evaluation standards for advertising and business cooperation.

[0107] In a preferred embodiment of the present invention, step 5 above, which analyzes the dynamic relationship between user social interaction and payment behavior based on content-differentiated feature vectors through an identity recognition association mechanism, and adjusts the advertising content matching strategy for users with strong psychological resistance to obtain advertising content matching results, may include:

[0108] Step 550: Using the identity association mechanism, analyze the correlation between differentiated content feature vectors and users' social behavior data to obtain the first analysis result representing the correlation between user identity and social behavior.

[0109] Step 551: Based on the first analysis results, predict the potential payment behavior patterns of users and generate payment behavior prediction results, including predictions of payment behavior tendencies.

[0110] Step 552: Based on the resistance level labels, identify and determine the target users marked as having prominent psychological resistance from the user group, and generate a target user identifier set;

[0111] Step 553: For the target users identified by the target user identifier set, generate an advertising strategy that is adjusted for the target users based on the payment behavior prediction results;

[0112] Step 554: Based on the adjusted advertising strategy, generate the final ad content matching results, including personalized ad content types and delivery timing configurations for different users.

[0113] In this embodiment of the invention, user social behavior data is extracted from the platform database. Social behavior includes a user's following relationships, comment content, sharing behavior, and interest groups joined within the platform. For example, user A follows 10 film and television accounts, is active in 3 interest groups related to suspense dramas, and shares an average of 2 short dramas to the social platform each week. For each user, a social behavior feature vector is constructed. Taking following relationships as an example, the types and number of accounts followed by the user are analyzed, and different weights are assigned to different types of accounts. Assuming that the weight of film and television accounts is 0.8, the weight of lifestyle accounts is 0.5, and the weight of game accounts is 0.3, and among the 10 accounts followed by user A, there are 6 film and television accounts, 3 lifestyle accounts, and 1 game account, then the feature value of their following relationship is: 6 × 0.8 + 3 × 0.5 + 1 × 0.3 = 4.8 + 1.5 +0.3=6.6. A correlation analysis is performed between the user's social behavior feature vector and content differentiation feature vector. For example, a certain dimension in the content differentiation feature vector represents the intensity of suspense elements. The analysis examines whether users who follow many suspense-related accounts also have high feature values ​​in that dimension. The correlation coefficient (such as the Pearson correlation coefficient) is calculated to measure the strength of the correlation. Assuming the correlation coefficient between the intensity of suspense elements and the number of suspense-related accounts followed is 0.75, it indicates a strong positive correlation. This analysis is performed on all users, resulting in a matrix where rows represent users and columns represent the correlation dimensions between content features and social behavior. Each element in the matrix represents the correlation strength value for the corresponding user in that correlation dimension. This is the first analytical result representing the correlation between user identity and social behavior.

[0114] Based on the first analysis results obtained in step 550, a paid behavior prediction mechanism is constructed, considering multiple factors, including the user's social behavior activity, content preference intensity, and payment history. For example, User B is very active in a science fiction interest group, and the content differentiation feature vector shows that their preference intensity for science fiction elements is 0.85. They also have two records of paying to watch science fiction short dramas in the past three months. For each user, their potential paid behavior tendency in various content types is calculated. Taking science fiction short dramas as an example, factors such as the user's science fiction content preference intensity, science fiction-related social behavior activity, and payment frequency are comprehensively considered. Assuming that the preference intensity weight is 0.4, the social behavior activity weight is 0.3, and the payment frequency weight is 0.3, User B's science fiction content preference intensity is 0.85, their science fiction-related social behavior activity (calculated by the number of times they speak in the science fiction group and the number of times they share science fiction content) is 0.9, and their payment frequency (number of payments in the past three months / total number of views) is 0.6. The potential payment tendency for users on science fiction short dramas is: 0.85×0.4+0.9×0.3+0.6×0.3=0.34+0.27+0.18=0.79. The above calculation is performed on all content types (such as suspense, romance, comedy, etc.) to generate a payment tendency vector for each user. Each element in the vector represents the user's potential payment tendency value on the corresponding content type, which is the payment behavior prediction result.

[0115] Based on users' resistance level tags, users marked as "high resistance" are selected from all users. Assuming the platform has 10,000 users, 1,500 of them have the "high resistance" tag, forming the target user set. To further confirm the degree of psychological resistance among these target users, their advertising interaction behavior details are analyzed. For example, for target user C, the number of ads they encountered in the past week, the percentage of ads skipped, and the frequency of closing ads are examined. If user C encountered 20 ads, skipped 18, and closed 15 in the past week, with a skip rate of 90% and a closing frequency of 75%, significantly higher than the platform average (assuming the platform average skip rate is 60% and the closing frequency is 40%), then user C is confirmed as a target user with high psychological resistance. This analysis is performed on all "high resistance" users to ultimately determine a precise target user identifier set, containing the IDs of all users confirmed to have high psychological resistance.

[0116] For users with a concentrated set of target user identifiers, the advertising strategy is adjusted based on their predicted payment behavior. Taking user D as an example, the predicted payment behavior results show that among suspense, romance, and comedy content, the potential payment tendency for suspense content is the highest at 0.82, followed by romance at 0.65, and comedy at the lowest at 0.4. Therefore, the proportion of comedy ads shown to user D is reduced, while the proportion of suspense ads is increased. For example, if the original ad distribution ratio was 30% suspense, 40% romance, and 30% comedy, it is adjusted to 50% suspense, 35% romance, and 15% comedy. At the same time, the timing of ad placement is adjusted. Based on user D's viewing habits, ads are placed during periods of higher activity. Assuming user D usually watches short dramas between 8 pm and 10 pm, the system will schedule 70% of the ads during this period. For different target users, personalized ad distribution ratios and timing configurations are developed based on their respective predicted payment behavior results, forming an advertising strategy tailored to the target user.

[0117] Based on the adjusted advertising strategy and combined with the user's content differentiation feature vector, the final ad content matching result is generated. Taking user E as an example, their content differentiation feature vector shows that their preference strength for plot tightness is 0.9, their preference strength for cast is 0.7, and their preference strength for special effects quality is 0.6. Ad content that matches user E's preferences is selected from the ad creative library. For example, ads with tight plots and strong casts are selected. For suspense ads, ad creatives with dramatic plots and excellent acting are selected. At the same time, according to the adjusted advertising strategy, the timing of ad placement is determined. Assuming that user E is more active on weekend evenings, relevant ads will be placed on weekend evenings. A detailed ad content matching result is generated for each user, including personalized ad content type for that user (such as suspense, romance, etc.), specific ad creative ID, and ad placement time window (such as Monday to Friday 19:00-21:00, weekend 20:00-22:00), etc.

[0118] By analyzing the correlation between content features and social behavior through identity association mechanisms, this approach achieves in-depth mining from users' external behavior to their internal identity. Traditional analysis only focuses on users' direct content preferences, while this step reveals users' deeper identity characteristics through social behavior data, such as follow relationships and interest groups. For example, it was discovered that although a user outwardly watches various types of short dramas, social behavior analysis revealed that their core identity is that of a "science fiction enthusiast," providing a more reliable basis for accurate recommendations. Based on the correlation analysis results, the accuracy of predicting user payment behavior patterns is improved. Combining multi-dimensional data such as users' social behavior activity and content preference intensity allows for a more comprehensive understanding of users' payment potential. For example, for users who are very active in specific interest groups but have not yet paid, this step can accurately predict their potential payment tendency, providing data support for marketing strategy adjustments and effectively improving paid conversion rates. Finally, it accurately identifies target users with strong psychological resistance. This approach avoids ineffective ad placement. Traditional advertising often ignores user resistance levels, applying the same strategy to all users, leading to aversion among some. This step, through resistance level tags and detailed analysis of ad interaction behavior, accurately identifies highly resistant users, reducing the randomness of ad placement. Adjusting ad strategies for target users effectively reduces psychological resistance by reducing ad types that users are not interested in and optimizing ad timing, making ads more aligned with user needs and habits. For example, for highly resistant users, ad placement times are adjusted to their prime time for watching short dramas, and the number of ads is reduced while ad quality is improved, increasing user acceptance and enhancing user experience. Personalized ad content matching results are generated, achieving precise and intelligent ad placement. Based on user content differentiation feature vectors and payment behavior predictions, ad content and placement plans are tailored to each user, improving the match between ads and user needs.

[0119] In a preferred embodiment of the present invention, step 6 above, based on the advertising content matching results, constructs a dynamic prediction mechanism for user payment behavior and generates a paid recommendation strategy; simultaneously, it establishes a closed-loop feedback channel for advertising performance, and adjusts psychological resistance assessment parameters, content quality assessment indicators, and payment prediction judgment logic by tracking the changing trends of advertising avoidance in real time, which may include:

[0120] Step 660: Based on the ad content matching results, extract user interaction data with the ads to form a user behavior dataset;

[0121] Step 661: Using the user behavior dataset, establish a dynamic rule system for predicting user payment behavior and obtain payment behavior prediction rules.

[0122] Step 662: Based on the paid behavior prediction rules, predict the paid behavior of the target users and generate a paid recommendation strategy, which specifically includes the type of recommended advertising content, the timing of recommendation, and the intensity of recommendation.

[0123] Step 663: Based on the paid recommendation strategy, establish a closed-loop feedback channel for ad performance, collect real-time data on user avoidance behavior towards ads that have been placed, and calculate the trend data of ad avoidance rate based on the avoidance behavior data.

[0124] Step 664: Based on the trend data, adaptively adjust the parameter values ​​in the psychological resistance assessment, the indicator items in the content quality assessment, and the judgment logic parameters of the paid behavior prediction rules.

[0125] In this embodiment of the invention, data is extracted from advertising delivery logs and user interaction records. The advertising delivery logs include information such as advertising delivery time, advertising type, and target audience (user ID). The user interaction records cover user actions such as clicking, viewing duration, skipping, and closing ads, along with corresponding timestamps. Taking user A as an example, in a certain advertising delivery, the following data was recorded: On December 10, 2024, at 19:30, a science fiction ad was delivered to user A. User A clicked skip after 5 seconds of ad playback. This record includes information such as time (19:30:05), action type (skip), and ad type (science fiction). The advertising interaction data of all users is filtered and organized, and invalid data, such as records where the ad exposure time is less than 1 second (considered as not being displayed normally), is removed. Then, the valid data is classified and collected according to user ID. Each user corresponds to a complete set of advertising interaction behavior records. These records together constitute a user behavior dataset. For example, after organization, user A's behavior dataset contains interaction records with 10 ads in the past week, detailing the time, behavior, and ad-related information for each interaction.

[0126] A dynamic rule system is constructed by combining hierarchical clustering and association rule mining. First, the behavioral features in the user behavior dataset are quantified. For example, ad viewing time is divided into different intervals: 0-5 seconds is scored as 1 point, 5-15 seconds as 2 points, and more than 15 seconds as 3 points; clicking is scored as 5 points, skipping as -3 points, and closing as -5 points. Next, users are hierarchically clustered. Taking user B as an example, in the past month's ad interaction data, ad viewing time was mostly between 0-5 seconds, and skipping occurred multiple times, resulting in a low overall score. Based on the scores of all users, users are divided into high interaction, medium interaction, and low interaction categories. The system identifies three user groups and then mines association rules within each group. For high-interaction groups, it is found that when users watch a certain type of advertisement for more than 15 seconds and click on related links, they are more likely to pay to watch that type of short drama. These associations are then refined into rules, such as "If a user watches a certain type of advertisement for 3 minutes and clicks on it (gets 5 points), then it is predicted that the user has a high probability of paying to watch this type of short drama." These rules are continuously updated and adjusted. Every time a certain amount of user behavior data is added (such as 1000 new records), clustering and rule mining are performed again to ensure that the rules can adapt to new user behavior patterns, forming a dynamic paid behavior prediction rule system.

[0127] Based on the paid behavior prediction rules established in step 661, paid behavior prediction is performed on target users. Taking user C as an example, their recent advertising interaction data is extracted. In the most recent suspense advertisement, the viewing time was 20 seconds (3 points), and the user clicked the link in the advertisement (5 points). According to the prediction rules, it is determined that user C has a high probability of paying to watch suspense short dramas. Personalized paid recommendation strategies are formulated for different users based on the prediction results. For users predicted to have a high probability of paying, such as user C, the recommendation strategy is to push paid discount activities for suspense short dramas, such as "first order 50% off". During the user's active period (analysis shows that user C is mostly active between 8-10 pm), the ad exposure frequency of this type of short drama is increased, and it is pushed 3-5 times a day. For users predicted to have a low probability of paying, such as user D, whose advertising interaction behavior is mainly to quickly skip, the strategy is adjusted to reduce direct paid recommendations and instead push free trial content of the type that the user may be interested in, attracting the user to gradually generate the willingness to pay. For example, a link to watch the first two episodes of a comedy short drama that user D once briefly watched is pushed to user D.

[0128] Establish a real-time data collection mechanism to collect user avoidance behavior data for displayed advertisements. Avoidance behavior includes actions such as skipping, closing, and swiping away. Taking user E as an example, on December 11, 2024, it was recorded that user E skipped 3 out of 5 advertisements received and closed 1. The advertisement avoidance rate was calculated using the formula: Advertisement Avoidance Rate = (Number of skips + Number of closes and other avoidance behaviors) ÷ Total number of advertisements received. User E's advertisement avoidance rate on that day was (3 + 1) ÷ 5 = 80%. The advertisement avoidance rate was statistically analyzed by time period (e.g., hourly, daily) and the trend was plotted. The advertisement avoidance rates of user E were recorded for 7 consecutive days on a daily basis as follows: 70%, 75%, 80%, 82%, 85%, 83%, and 80%. By analyzing this data, it was found that user E's ad avoidance rate showed an upward trend from day 3 to day 5, followed by a slight decrease. These trend data were saved and analyzed. At the same time, the avoidance rate trends of different user groups and different ad types were compared. For example, it was found that the avoidance rate of game ads among young users increased on weekends, while the avoidance rate of life service ads among middle-aged and elderly users remained relatively stable.

[0129] Based on the trend data of ad avoidance rates, the parameters for assessing psychological resistance are adjusted. For example, if it is found that the avoidance rate of a certain type of ad for a certain user group is continuously increasing, it indicates that the psychological resistance of this group to such ads is strengthening. Therefore, the weight of relevant indicators in the psychological resistance assessment algorithm for this type of ad will be increased. For instance, the weight of ad skipping behavior in the assessment of this type of ad will be increased from 0.4 to 0.6, making the calculation of the degree of psychological resistance for this group more sensitive. Regarding content quality assessment indicators, if the avoidance rate of ads related to a certain type of content increases and the user payment conversion rate decreases, assessment indicators will be added or adjusted. For example, for short videos with dragging plots... For drama genres, the original content quality assessment did not consider plot pacing indicators. Now, "plot pacing smoothness" will be included in the assessment indicator system and given a weight of 0.2. The score of this indicator will be calculated through user comment analysis, viewing interruption data, and other methods. The judgment logic parameters of the paid behavior prediction rules will also be adjusted. If it is found that the accuracy of the current rule in predicting user paid behavior has decreased, such as the actual payment rate of users who were originally predicted to have a high probability of paying being less than 30%, the thresholds and conditions in the rule will be adjusted. For example, the score threshold for triggering a high probability of payment prediction in a certain rule will be lowered from 8 points to 6 points to expand the prediction range and improve the accuracy of the prediction.

[0130] By extracting and organizing user interaction data with advertisements, a comprehensive and accurate user behavior dataset was constructed. Compared to traditional, fragmented data collection methods, this step ensures data integrity and validity. The removal of invalid data prevents erroneous data from interfering with subsequent analysis, providing a reliable data foundation for accurate paid behavior prediction and strategy development. A dynamic paid behavior prediction rule system was established, enabling predictions to adapt to constantly changing user behavior patterns. Traditional static rules struggle to cope with the diversity and dynamism of user behavior, while this step, through hierarchical clustering and association rule mining combined with a real-time update mechanism, can promptly capture new user behavioral characteristics and patterns, improving paid behavior prediction accuracy. The accuracy and reliability of behavioral predictions provide strong support for the platform to formulate effective marketing strategies. Personalized paid recommendation strategies are generated based on the prediction results, and differentiated strategies are developed according to the different users' payment potential. This not only improves the conversion rate of high-potential users but also cultivates the willingness to pay among low-potential users through guiding strategies, thereby increasing the platform's commercial revenue and avoiding resource waste. Real-time tracking of ad avoidance rate trends provides the platform with timely and intuitive feedback on advertising effectiveness. By analyzing avoidance rates for different user groups and ad types, problems in ad placement can be quickly identified, enabling the platform to optimize ad placement strategies in a timely manner and improve the efficiency and effectiveness of ad placement.

[0131] like Figure 2 As shown, embodiments of the present invention also provide a short drama user behavior analysis system, comprising:

[0132] The data acquisition module is used to collect dynamic behavioral data of users on the short drama platform, including real-time viewing behavior data and advertising interaction data, and to establish a time-series database of user behavior.

[0133] The behavior construction module is used to extract three fixed detection points based on the user behavior time series database, construct the user behavior feature topology relationship with the three fixed detection points as vertices, analyze the user behavior feature topology relationship, and generate behavior feature adjustment values.

[0134] The resistance assessment module is used to dynamically identify users' preferences in entertainment, information, social and emotional needs based on behavioral characteristic adjustment values. At the same time, it constructs a psychological resistance assessment mechanism to quantify the degree of users' psychological resistance to advertising and dynamically update the resistance level label.

[0135] The differentiated evaluation module is used to quantify the plot structure, performance artistry, and production technical parameters of short dramas based on resistance level tags, and to build a dynamic evaluation matrix by combining paid conversion rate to generate content differentiation feature vectors.

[0136] The matching strategy module is used to analyze the dynamic relationship between users' social and payment behaviors based on differentiated feature vectors and identity recognition association mechanisms, and to adjust the advertising content matching strategy for users with strong psychological resistance in order to obtain advertising content matching results.

[0137] The paid prediction module is used to build a dynamic prediction mechanism for user payment behavior based on the ad content matching results and generate paid recommendation strategies. At the same time, it establishes a closed-loop feedback channel for ad performance, and adjusts the psychological resistance assessment parameters, content quality assessment indicators and paid prediction judgment logic by tracking the changing trend of ad avoidance in real time.

[0138] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0139] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0140] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0141] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing user behavior in short dramas, characterized in that, The method includes: Step 1: Obtain dynamic behavioral data of users on the short drama platform through multi-dimensional data, including real-time viewing behavior data and advertising interaction data, and establish a time-series database of user behavior. Step 2: Based on the user behavior time-series database, extract three fixed detection points, construct a user behavior feature topology using these three fixed detection points as vertices, and analyze the user behavior feature topology to generate behavior feature adjustment values. Specifically, this includes: extracting the user's starting behavior point, advertising touchpoint, and ending behavior point during short drama viewing from the user behavior time-series database as three fixed detection points; constructing user behavior connection relationships with these three fixed detection points as vertices based on their order in the user behavior time-series; using these user behavior connection relationships, analyzing the behavior transition frequency between vertices in the graph to determine the behavior association strength, and calculating the time interval distribution between adjacent vertices to extract time interval features; combining the behavior association strength and time interval features, and generating behavior feature adjustment values ​​representing user behavior patterns through weighted fusion calculation. Step 3: Based on behavioral characteristic adjustment values, dynamically identify user preferences in entertainment, information, social, and emotional needs. Simultaneously, construct a psychological resistance assessment mechanism to quantify the degree of user psychological resistance to advertising and dynamically update resistance level labels. Specifically, this includes: based on behavioral characteristic adjustment values, identifying and quantifying the intensity of user preference tendencies in entertainment, information, social, and emotional needs dimensions using preset mapping rules; based on preference intensity, extracting key indicators from advertising interaction data, including ad skip rate and ad closing frequency, and calculating user psychological resistance scores using a predefined psychological resistance assessment algorithm; based on user psychological resistance scores, classifying user psychological resistance into multiple levels according to preset score range thresholds, and assigning corresponding initial resistance level labels to users; monitoring new user behavioral characteristics and advertising interaction data, repeatedly performing need identification, psychological resistance calculation, and level classification, and dynamically updating user resistance level labels based on the latest results. Step 4: Based on the resistance level tags, quantify the short drama's plot structure, performance artistry, and production technical parameters, and construct a dynamic evaluation matrix in conjunction with the paid conversion rate to generate a content differentiation feature vector. Specifically, this includes: calculating the plot coherence quantification index based on user interruption behavior data in the resistance level tags; calculating the performance attractiveness score based on user interaction behavior data; generating production technical parameters in conjunction with production specification data; spatiotemporally aligning the plot coherence quantification index, performance attractiveness score, and production technical parameters with the real-time paid conversion rate dataset; dynamically weighting and fusing the aligned multi-dimensional data according to a preset time window to construct a dynamic evaluation matrix where rows and columns respectively represent content attributes and time series; performing principal component decomposition on the dynamic evaluation matrix to determine the main feature combinations with relatively prominent eigenvalues, and normalizing the extracted feature combinations to generate a feature vector representing the content differentiation of the short drama. Step 5: Based on the content differentiation feature vector, analyze the dynamic relationship between user social interaction and payment behavior through the identity association mechanism, and adjust the advertising content matching strategy for users with strong psychological resistance to obtain advertising content matching results. Specifically, this includes: using the identity association mechanism to analyze the correlation between the differentiated content feature vector and users' social behavior data, obtaining a first analysis result representing the correlation between user identity and social behavior; based on the first analysis result, predicting potential payment behavior patterns of users, generating payment behavior prediction results including predictions of payment behavior tendencies; identifying and determining target users marked as having strong psychological resistance from the user group based on resistance level labels, generating a target user identifier set; for the target users identified by the target user identifier set, generating an advertising strategy adjusted for the target users based on the payment behavior prediction results; and based on the adjusted advertising strategy, generating the final advertising content matching results, including personalized advertising content types and delivery timing configurations for different users. Step 6: Based on the ad content matching results, construct a dynamic prediction mechanism for user payment behavior and generate a paid recommendation strategy; at the same time, establish a closed-loop feedback channel for ad performance, and adjust the psychological resistance assessment parameters, content quality assessment indicators and payment prediction judgment logic by tracking the changing trend of ad avoidance in real time.

2. The short drama user behavior analysis method according to claim 1, characterized in that, Based on ad content matching results, a dynamic prediction mechanism for user payment behavior is constructed to generate a paid recommendation strategy. Simultaneously, a closed-loop feedback channel for ad performance is established. By tracking real-time trends in ad avoidance, the psychological resistance assessment parameters, content quality assessment indicators, and payment prediction judgment logic are adjusted, including: Based on the ad content matching results, user interaction data with ads is extracted to form a user behavior dataset. By utilizing user behavior datasets, a dynamic rule system for predicting user payment behavior is established to obtain payment behavior prediction rules; Based on paid behavior prediction rules, the paid behavior of target users is predicted, and a paid recommendation strategy is generated. Based on the paid recommendation strategy, a closed-loop feedback channel for ad performance is established to collect data on user avoidance behavior towards ads in real time, and based on the avoidance behavior data, the trend data of ad avoidance rate is calculated. Based on trend data, we adaptively adjust the parameter values ​​in psychological resistance assessment, the indicator items in content quality assessment, and the judgment logic parameters of the paid behavior prediction rules.

3. The short drama user behavior analysis method according to claim 2, characterized in that, The paid recommendation strategy includes the type of recommended ad content, the timing of recommendation, and the intensity of recommendation.

4. A short drama user behavior analysis system, wherein the system implements the method as described in any one of claims 1 to 3, characterized in that, include: The data acquisition module is used to collect dynamic behavioral data of users on the short drama platform, including real-time viewing behavior data and advertising interaction data, and to establish a time-series database of user behavior. The behavior construction module is used to extract three fixed detection points based on the user behavior time series database, construct the user behavior feature topology relationship with the three fixed detection points as vertices, analyze the user behavior feature topology relationship, and generate behavior feature adjustment values. The resistance assessment module is used to dynamically identify users' preferences in entertainment, information, social and emotional needs based on behavioral characteristic adjustment values. At the same time, it constructs a psychological resistance assessment mechanism to quantify the degree of users' psychological resistance to advertising and dynamically update the resistance level label. The differentiated evaluation module is used to quantify the plot structure, performance artistry, and production technical parameters of short dramas based on resistance level tags, and to build a dynamic evaluation matrix by combining paid conversion rate to generate content differentiation feature vectors. The matching strategy module is used to analyze the dynamic relationship between users' social and payment behaviors based on differentiated feature vectors and identity recognition association mechanisms, and to adjust the advertising content matching strategy for users with strong psychological resistance in order to obtain advertising content matching results. The paid prediction module is used to build a dynamic prediction mechanism for user payment behavior based on the ad content matching results and generate paid recommendation strategies. At the same time, it establishes a closed-loop feedback channel for ad performance, and adjusts the psychological resistance assessment parameters, content quality assessment indicators and paid prediction judgment logic by tracking the changing trend of ad avoidance in real time.

5. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 3.

Images