Cloud big data real-time flow casting method and system based on intelligent analysis

Abstract

The invention provides a cloud big data real-time streaming method and system based on intelligent analysis, and the method comprises the steps: obtaining a standardized behavior data set through the collection of the real-time interaction behaviors of a user, including the movie watching time length and member page interaction records, and formatting the real-time interaction behaviors; identifying watching duration and image quality preference through the standardized behavior data set, determining a user traffic consumption type according to the watching duration and the image quality preference, and obtaining a user package adaptation type; according to the optimized package combination scheme, the display sequence of the traffic card packages on the member page is adjusted, and according to the click frequency, the top package scheme is set to obtain a personalized package traffic investment scheme; the method comprises the following steps: acquiring a user package click conversion rate from a personalized package flow scheme, and identifying a user preference change trend to determine recommendation weight configuration by continuously fusing newly collected watching duration;

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a cloud big data real-time streaming method and system based on intelligent analysis. BACKGROUND

[0002] The operation mode of combining short video content consumption with traffic service is increasingly important in the field of digital marketing, especially in the scenario of combining short drama membership promotion with traffic card sales. This mode promotes the dual improvement of user stickiness and business conversion by precisely meeting users' content viewing and network needs. However, the existing methods in this field often face significant technical challenges in dealing with user diversification needs and real-time decision-making, especially in how to effectively use user behavior data to achieve precise recommendation. Existing methods usually recommend fixed traffic card packages based on users' historical consumption records. However, this approach is difficult to adapt to the dynamic changes of users in short drama viewing, such as from occasional viewing of short dramas to high-frequency viewing, or from standard definition to high-definition viewing. The limitation of this static recommendation is that it cannot capture the real-time evolution of user behavior patterns, resulting in a mismatch between recommended traffic card packages and actual user needs, reducing purchase conversion rates. In addition, in the promotion of short drama membership and traffic card sales, user behavior data in short drama viewing, such as viewing time, episode number, quality preference, and daily active period, are highly heterogeneous and dynamic. For example, a user may initially choose a small-capacity monthly card, but as the frequency of watching high-definition short dramas increases, their traffic needs may quickly shift to large-capacity quarterly or annual cards. This dynamic change in demand requires real-time optimization of traffic card package display. Traditional recommendation systems often rely on fixed rules or historical data and cannot dynamically adjust the displayed traffic card combination based on real-time interaction behaviors such as users' dwell time, click frequency, and other real-time interaction behaviors on the membership page. For example, a user repeatedly views the details page of a large-capacity traffic card on the membership page, but ultimately does not purchase, possibly because the display method or combination scheme of the package fails to effectively stimulate their purchase intention. The combination of analysis of this real-time interaction data and optimization of the package requires the system to have the ability to respond quickly and adjust dynamically, and the shortcomings of existing technologies in this regard result in a decrease in the matching degree of the recommendation effect and user needs. Therefore, how to dynamically adjust the display strategy and combination scheme of the traffic card package based on real-time analysis of user short drama viewing behavior and membership page interaction data has become a key problem in the scenario of combining short video content consumption with traffic service. SUMMARY

[0003] The present application provides a cloud big data real-time streaming method based on intelligent analysis, mainly including: By collecting real-time user interaction behavior and formatting it, a standardized behavior dataset is obtained. This dataset includes short drama viewing time and member page interaction records. The standardized behavior dataset is used to identify user viewing time and image quality preferences, and based on these preferences, the user's data plan suitability type is determined. From the user's data plan suitability type, member page dwell time and data plan card click frequency are extracted to obtain the member page browsing path. This is then matched with the data plan preferences of similar user groups to generate a preliminary data plan recommendation list. Based on the member page browsing path, user page navigation behavior is analyzed to identify user behavior patterns, and real-time interaction is evaluated based on these patterns. The interaction is used to match the user's package type, sort the initial package recommendation list, and determine the optimized package combination scheme. Based on the optimized package combination scheme, the display order of data SIM card packages on the member page is adjusted, and packages are prioritized based on click frequency to generate a personalized package traffic distribution scheme. The user package click-through rate is extracted from the personalized package traffic distribution scheme, and combined with newly collected viewing time to identify user preference trends and determine the recommendation weight configuration. Based on the recommendation weight configuration, behavior pattern recognition is reprocessed, and based on the matching degree between dynamic demand changes and real-time interactive behavior, the final data SIM card package recommendation result is output, including package type, data capacity, and price tier.

[0004] Furthermore, the process of collecting real-time user interaction behavior and formatting it to obtain a standardized behavior dataset includes: The system acquires real-time operation records of users on the short drama player and interaction records on the membership page, extracts short drama viewing duration, number of pauses, video quality switching records, package card click coordinates, and page dwell time, and aligns them according to timestamps to form an original behavior sequence; abnormal records are removed from the original behavior sequence, the data format is converted, and the numerical range is unified to the 0 to 1 interval through a normalization method to generate the standardized behavior dataset.

[0005] Furthermore, the process of identifying user viewing time and image quality preferences through a standardized behavioral dataset, and determining the user's package suitability type based on viewing time and image quality preferences, includes: Viewing duration sequences and image quality selection records are extracted from the standardized behavior dataset. The ratio of the daily cumulative viewing duration to the preset duration is calculated as a viewing density index. The percentage of standard definition, high definition, and ultra-high definition selections is statistically analyzed to form an image quality preference distribution. These are combined to form a user viewing behavior feature vector. Based on the user viewing behavior feature vector, the unit time data consumption is calculated by combining the image quality selection frequency and bitrate standard. Based on the viewing density index and image quality preference distribution, the high data consumption type, high-speed data consumption type, or low data consumption type is determined, and the corresponding package capacity is matched to generate the user package adaptation type.

[0006] Further, the member page stay duration and package card click frequency are extracted from the user package adaptation type, the member page browsing path is obtained, the traffic card package preference of the similar user group is matched, and a preliminary package recommendation list is generated, including: The member page stay duration and package card click frequency are extracted from the user package adaptation type, the page jump path is recorded, and the user page interaction behavior data set is generated; the package attention value is calculated according to the user page interaction behavior data set, the group preference value is calculated in combination with the historical package purchase records of the similar user group, the comprehensive recommendation score is calculated, and the preliminary package recommendation list is sorted to generate the optimized package combination scheme.

[0007] Further, the user page jump behavior is analyzed according to the member page browsing path, the user behavior mode is identified, the matching degree of real-time interaction behavior and user package adaptation type is evaluated according to the user behavior mode, the preliminary package recommendation list is sorted, and the optimized package combination scheme is determined, including: The page jump sequence is extracted according to the member page browsing path, the interaction frequency index and the repeated access times of the package card are calculated, and the user page interaction feature set is generated; the high-frequency interaction mode is determined according to the user page interaction feature set, and the package access sequence and the stay duration are recorded; the coincidence ratio of the real-time access package type and the user package adaptation type is calculated as the consistency score, the package priority value is given based on the consistency score, the preliminary package recommendation list is sorted, and the optimized package combination scheme containing the main promotion and the alternative package is generated.

[0008] Further, the display order of the traffic card package in the member page is adjusted according to the optimized package combination scheme, the package is topped up according to the click frequency, and a personalized package streaming scheme is generated, including: The recommendation priority value is extracted from the optimized package combination scheme, the comprehensive display weight is calculated in combination with the user historical click times, and the package display order is determined; the page display parameters are configured according to the comprehensive display weight, the display area and the visual effect of the top package, the medium weight package and the low weight package are set, and the personalized package streaming scheme containing the position sorting and the style is generated.

[0009] Further, the display order of the traffic card package in the member page is adjusted according to the optimized package combination scheme, the package is topped up according to the click frequency, and a personalized package streaming scheme is generated, including: The recommendation weight value and the user matching degree score are extracted from the optimized package combination scheme, the visual highlighting coefficient is calculated in combination with the user historical click times and the stay duration, and the package visual configuration parameter set is generated; the package display area, the color saturation and the border style are adjusted according to the package visual configuration parameter set, the page grid number is distributed, the dynamic effect is added, and the personalized page display layout containing the position, the size and the style is generated.

[0010] Further, the user package click conversion rate is extracted from the personalized package investment scheme, the change trend of user preference is identified in combination with newly collected viewing time, and a recommendation weight configuration is determined, including: The number of clicks and the number of purchases are extracted from the personalized package investment scheme, the click conversion rate is calculated, the time series data set is formed in combination with the latest viewing time data, the preference change point is detected according to the time series data set, the change trend is counted, the recommendation weight of the large-flow package or the light package is adjusted, and the recommendation weight configuration is generated.

[0011] Further, the behavior pattern recognition is reprocessed according to the recommendation weight configuration, the matching degree of dynamic demand change and real-time interaction behavior is output, and the final traffic card package recommendation result is output, including the package type, the traffic capacity and the price level, including: According to the recommendation weight configuration, the behavior pattern recognition threshold is updated in combination with the user click and purchase behavior data, the behavior pattern identification is regenerated, the dynamic demand feature is extracted according to the behavior pattern identification, the traffic consumption change rate and the number of package detail page visits are calculated, and the matching degree value is determined; the matching degree value is extracted, the package type, the traffic capacity value and the price level are extracted, and the structured final traffic card package recommendation result is generated.

[0012] The application provides a cloud big data real-time investment system based on intelligent analysis, mainly including: The behavior data acquisition module is used for acquiring real-time interaction behavior of users, including short drama viewing time and member page interaction record, and standardized behavior data set is obtained by formatting processing; The user preference identification module is used for identifying viewing time and picture quality preference through the standardized behavior data set, determining the user traffic consumption type according to the viewing time and the picture quality preference, and obtaining the user package adaptation type; The preliminary recommendation generation module is used for extracting the member page stay time and the package card click frequency statistics from the user package adaptation type, simultaneously acquiring the member page browsing path, matching the traffic card package preference of similar user groups, and obtaining the preliminary package recommendation list; The behavior pattern analysis module is used for analyzing user page jump behavior according to the member page browsing path, identifying the current user behavior pattern, if the user behavior pattern is high-frequency interaction, the matching degree of real-time interaction behavior and user package adaptation type is evaluated, the preliminary package recommendation list is prioritized, and the optimized package combination scheme is determined; A personalized flow injection module is used to adjust the display order of the flow card package on the member page according to the optimized package combination scheme, and the package scheme is topped according to the click frequency to obtain a personalized package flow injection scheme; a weight configuration updating module is used to obtain the user package click conversion rate from the personalized package flow injection scheme, and the recommendation weight configuration is determined by continuously integrating the newly collected viewing time to identify the user preference change trend; A dynamic recommendation output module is used to reprocess the behavior pattern recognition by using a feedback mechanism for the recommendation weight configuration, and if the dynamic demand change and the real-time interactive behavior matching degree are improved, a final flow card package recommendation output is obtained, including the package type, the flow capacity, the price level and the recommendation reason.

[0013] The technical scheme provided by the embodiment of the application can include the following beneficial effects: The application discloses a cloud big data real-time flow injection method and system based on intelligent analysis, and aims at the accurate matching problem of user flow consumption demand in a short drama watching scene, constructs a standardized behavior data set by collecting user short drama watching time, picture quality preference and member page interactive data, analyzes user flow consumption types and forms a package adaptation classification system. The application identifies the sensitivity and demand intensity of users to flow consumption by cross analyzing the watching time density and the picture quality preference, and induces the flow consumption types into large flow, light weight and high speed consumption types, dynamically adjusts the package recommendation weight and display order by combining the member page browsing path and the interactive frequency, optimizes the page layout to highlight the high matching degree package. The application dynamically updates the recommendation weight by continuously integrating the user click conversion rate and the preference change trend by using a feedback mechanism, and finally outputs a personalized recommendation scheme including the package type, the flow capacity, the price and the recommendation reason. The application significantly improves the accuracy of the flow card package recommendation and the user conversion rate, and optimizes the resource configuration efficiency in the short drama scene. BRIEF DESCRIPTION OF DRAWINGS

[0014] Fig. 1 It is a flow chart of the cloud big data real-time flow injection method based on intelligent analysis.

[0015] Fig. 2 It is a structural schematic diagram of the cloud big data real-time flow injection system based on intelligent analysis. DETAILED DESCRIPTION

[0016] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described in the specification below in conjunction with the drawings in the specification. Obviously, the described embodiments are only some of the embodiments of the specification, not all. Based on the embodiments in the specification, all other embodiments obtained by those of ordinary skill in the art without creative labor should belong to the scope of protection of the specification.

[0017] As Figs. 1-2 The embodiment of the cloud big data real-time streaming method and system based on intelligent analysis can specifically include: S101, by collecting user real-time interaction behavior, including short play watching time and member page interaction record, formatting processing obtains standardized behavior data set.

[0018] Obtain the real-time operation record of the user in the short play player, extract the watching time, pause times and quality switching record from the player log, and collect the click coordinates and page staying time of the package card from the member page, align the time sequence according to the time stamp of each data source, and obtain the original behavior sequence containing the playing behavior and page interaction. Data cleaning is performed on the original behavior sequence, and abnormal records with buffer time longer than the preset threshold are removed, various behavior data is converted according to the JSON format specification, the numerical range is unified to the interval of 0 to 1 through the maximum and minimum value normalization method, and the standardized behavior data set is determined.

[0019] In an embodiment, when obtaining the user real-time interaction behavior, the short play player log record contains the Unix time stamp, video ID, watching time seconds, pause times count and quality switching record of each playing operation, wherein the quality switching record marks the specific time when switching from standard definition 360P to high definition 720P or super definition 1080P. The package card click coordinates of the member page record the touch position of the user on the screen, and the page staying time is timed from the page loading until the user jumps or exits. Time sequence alignment uses a unified time base to convert the time stamps of different data sources into millisecond-level time stamps in the same time zone, and arranges them in time sequence to form a continuous behavior sequence.

[0020] Exemplarily, when the user switches to the member page to browse the package after watching the short drama for 10 minutes, the original behavior sequence records the watching behavior and page interaction behavior in chronological order. In the data cleaning process, the preset threshold of the buffering duration is set to 30 seconds, and the records exceeding the threshold are marked as invalid data caused by network anomalies. The JSON format specification conversion maps various behavior data into a key-value pair structure, the watching duration is mapped into the duration field, and the click coordinates are mapped into the position field. The normalization processing adopts the maximum and minimum value method, maps the watching duration from the range of 0 to 3600 seconds to the interval of 0 to 1, and ensures that the data of different dimensions are comparable.

[0021] S102, identify the watching duration and the picture quality preference from the standardized behavior data set, determine the user traffic consumption type according to the watching duration and the picture quality preference, and obtain the user package adaptation type.

[0022] The watching duration and the picture quality selection record are extracted from the standardized behavior data set, the daily watching duration cumulative value is divided by 24 hours to obtain a watching density index, the proportion of the number of times that the user selects standard definition, high definition and ultra definition is counted to determine the picture quality preference distribution, and the watching density index and the picture quality preference distribution are combined to form a user watching behavior feature vector. The traffic consumption of the user watching behavior feature vector is evaluated, the picture quality selection frequency is multiplied by the corresponding code rate to obtain the unit time traffic consumption according to the code rate standard that standard definition corresponds to 500 kbps, high definition corresponds to 2 Mbps, and ultra definition corresponds to 4 Mbps, if the watching density index exceeds 0.3 and the proportion of ultra definition selection exceeds 0.6, it is determined as a large traffic consumption type, if the watching is concentrated in the evening from 8 to 11 o'clock, it is determined as a high-speed traffic consumption type, and the rest is determined as a light traffic consumption type. The package capacity is matched according to the traffic consumption type, the large traffic consumption type corresponds to a monthly package of more than 30 GB, the high-speed traffic consumption type corresponds to a 20 GB package containing idle time traffic, and the light traffic consumption type corresponds to a 10 GB basic package, and the user package adaptation type is determined.

[0023] In an embodiment, when the user watching behavior features are extracted from the standardized behavior data set, the watching density index reflects the immersion degree of the user to the short drama content.

[0024] Specifically, the cumulative value of the watching duration of the user within 24 hours is obtained, which is divided by 24 to obtain the hourly mean value as the watching density basic value, and then the continuity of the watching period is adjusted by weighting, the weight of the period with continuous watching for more than 30 minutes is set to 1.2, and the weight of the period with intermittent watching is set to 0.8, to obtain the watching density index. The picture quality selection record includes the time point of each time the user actively switches the picture quality and the selected picture quality level, the proportion of the number of selections of each picture quality level to the total watching time is counted to form the picture quality preference distribution.

[0025] It should be noted that the user viewing behavior feature vector is composed of two parts of viewing density index and picture quality preference distribution, wherein the viewing density index is a single numerical value, and the picture quality preference distribution is a three-dimensional vector, respectively corresponding to the selection proportion of standard definition, high definition and ultra definition.

[0026] Preferably, the traffic consumption evaluation process determines the code rate corresponding to different picture qualities according to the video encoding standard. When the standard definition content is encoded by H.264, the code rate is about 500kbps, the code rate of high definition content is increased to 2Mbps, and the code rate of ultra definition content reaches 4Mbps. The average traffic consumption rate is obtained by multiplying the viewing time of the user under each picture quality by the corresponding code rate, and then dividing the total viewing time. When the viewing density index exceeds 0.3, it indicates that the user watches more than 7 hours per day, and at the same time, the proportion of ultra definition selection exceeds 0.6, which indicates that the user prefers high picture quality content. At this time, it is determined as a large traffic consumption type, and the monthly traffic demand is expected to exceed 30GB.

[0027] Exemplarily, for the user whose viewing period is concentrated in the evening from 8pm to 11pm, although the total viewing time may not be long, the bandwidth requirement is high during the network peak period, and it is determined as a high-speed traffic consumption type, and a 20GB package containing idle time traffic is suitable for recommendation, and the idle time traffic at night is unlimited speed.

[0028] In one possible implementation, the viewing behavior of the user of the light traffic consumption type presents a fragmented feature, the single viewing time is usually within 15 minutes, the picture quality of standard definition is often selected to save traffic, and the monthly traffic demand is within 10GB, which can be matched with the basic package to meet the demand.

[0029] The short drama viewing time data and picture quality selection records are extracted from the standardized behavior data set, the user viewing time density and picture quality preference tendency are analyzed, the viewing time and picture quality selection are cross-analyzed, the sensitivity and demand intensity of the user to traffic consumption are identified, the user traffic usage behavior is summarized into a large traffic consumption type, a light traffic consumption type and a high-speed traffic consumption type according to the frequency of viewing behavior and the high-low degree of picture quality requirement, the corresponding relationship between user traffic consumption behavior and package demand is established, and a user package adaptation type classification system is formed.

[0030] The short drama viewing duration sequence and the picture quality selection record are extracted from the standardized behavior data set, 24 hours are divided into 24 time periods, the viewing times and the cumulative duration in each time period are counted, a duration density matrix is constructed, taking the time period as the row, the viewing times and the duration as the column, and the selection frequency proportion of three picture quality levels, namely, standard definition, high definition and ultra definition, and the switching sequence are counted, so as to obtain a user viewing behavior basic feature set. Cross feature extraction is performed on the user viewing behavior basic feature set, the peak time period in which the viewing times exceed the average value is identified from the duration density matrix, the ratio of the high picture quality selection frequency in the peak time period to the total selection times in the time period is calculated as the traffic sensitivity, the proportion of the total high definition and ultra definition viewing duration to the total viewing duration is defined as the demand intensity index, and a user traffic consumption feature vector is obtained. According to the user traffic consumption feature vector, the consumption behavior is classified, if the demand intensity index exceeds 0.7 and the daily average viewing duration exceeds 180 minutes, the user is classified as a large traffic consumption type, if the peak time period is concentrated in the evening and the picture quality switching frequency exceeds 3 times per hour, the user is classified as a high-speed traffic consumption type, and if the daily average viewing duration is less than 60 minutes and the standard definition selection proportion exceeds 0.5, the user is classified as a light traffic consumption type, and the user traffic usage behavior category is determined. A package demand mapping is established based on the user traffic usage behavior category, the large traffic consumption type corresponds to a monthly package of more than 30 GB, the high-speed traffic consumption type corresponds to a 20 GB peak speed package, and the light traffic consumption type corresponds to a 10 GB basic package, and a user package adaptation type classification system containing the consumption type identifier and the package specification is generated according to the mapping relationship.

[0031] In an embodiment, when constructing the duration density matrix, 24 hours of a day are divided into 24 independent time periods, and each time period corresponds to a row of the matrix.

[0032] Specifically, the first column of the matrix records the number of times that the user opens the short drama player in the time period, and the second column records the cumulative viewing duration in the time period, in minutes. When the user opens the player 3 times between 2 a.m. and 3 a.m., and the cumulative viewing duration is 45 minutes, the value of the 3rd row of the matrix is [3, 45]. Through data accumulation for 7 consecutive days, the average value of each time period is calculated to form a stable duration density matrix. The picture quality selection record contains the timestamp of each time that the user actively switches the picture quality, the picture quality level before switching and the picture quality level after switching. Statistics show that the frequency of switching from standard definition to high definition accounts for 35% of the total switching times, the frequency of switching from high definition to ultra definition accounts for 45%, and the remaining 20% is the operation of reducing the picture quality. Such switching sequence reflects the progressive demand of the user for the picture quality.

[0033] It should be noted that the user viewing behavior base feature set is a multi-dimensional data structure, including a time length density matrix, a picture quality selection frequency distribution, and a picture quality switching sequence, which depict the user's viewing habits from different angles. In the cross-feature extraction process, the identification of the peak period is based on the statistical analysis of the viewing times column in the time length density matrix. The mean and standard deviation of the viewing times of 24 time periods are calculated, and when the viewing times of a certain time period exceed the mean plus one standard deviation, it is marked as a peak period.

[0034] For example, the viewing times from 20:00 to 22:00 are 8 and 7 respectively, and the average viewing times throughout the day are 3, with a standard deviation of 2. The viewing times of these two periods both exceed the threshold of 5, so they are identified as peak periods. The calculation of traffic sensitivity focuses on the picture quality selection behavior in the peak period. If the number of times the user selects high definition or ultra definition in the peak period accounts for 80% of the total selection times in the period, the traffic sensitivity value is 0.8, indicating that the user has a higher requirement for picture quality in the high-frequency viewing period.

[0035] In one possible implementation, the demand intensity index takes into account the overall preference of the user for high-quality content. The total viewing time of high-definition content and the total viewing time of ultra-clear content are added and divided by the total viewing time to obtain the base value of the demand intensity. If a user watches 100 minutes of standard definition content, 300 minutes of high-definition content, and 200 minutes of ultra-clear content in a week, the demand intensity index is 500 divided by 600, i.e. 0.83, indicating that the user has a strong demand for high-quality.

[0036] By way of example, the user traffic consumption feature vector is a two-dimensional vector containing traffic sensitivity and demand intensity index, and the two dimensions together determine the user's traffic consumption type. Large traffic consumption type users exhibit sustained high-intensity viewing behavior, with a demand intensity index exceeding 0.7 indicating that more than 70% of the viewing time selects high definition or ultra definition picture quality, and a daily viewing time exceeding 180 minutes equivalent to watching more than 6 episodes of short drama content per day. Such users usually have viewing behavior in multiple periods and are not limited to specific times, and their demand for traffic presents an all-weather feature. The characteristic of high-speed traffic consumption type users is that the viewing behavior is concentrated in the network peak period, usually from 19:00 to 23:00, which is a period of serious network congestion. Users frequently switch between different picture qualities to seek a smooth viewing experience, with more than 3 switches per hour indicating that the user's sensitivity to network speed is higher than his insistence on picture quality. The viewing behavior of light traffic consumption type users is relatively conservative, with a daily viewing time of less than 60 minutes usually watching only 2 to 3 episodes of short drama, and a standard definition selection ratio of more than 50% indicating that the user pays more attention to saving traffic than pursuing high-quality experience.

[0037] For example, the demand intensity index of a certain user is 0.75, and the average daily viewing time is 200 minutes. According to the classification rule, it is determined as a large flow consumption type, and a monthly package of more than 30GB is recommended, which contains enough high-speed flow to meet the all-weather high-definition viewing needs.

[0038] In an embodiment, the establishment of the package demand mapping relationship considers the actual flow usage characteristics of users of different consumption types. The large flow consumption type user calculates according to an average of 15 minutes per short drama, 4Mbps code rate for super-definition quality, consumes about 450MB flow per set, and needs 5.4GB flow per day to watch 12 sets, and the monthly demand exceeds 160GB, but considering the WiFi usage scene, the actual mobile network flow demand is about 30GB. Although the high-speed flow consumption type user does not require as much total flow as the large flow user, the network quality requirement for a specific period is higher, and the 20GB peak speed package provides an exclusive bandwidth channel during the evening peak period to ensure a smooth viewing experience.

[0039] It can be understood that the user package adaptation type classification system realizes accurate product recommendation through the combination of consumption type identification and package specifications, and improves the efficiency of user selection of appropriate packages.

[0040] S103, extract the member page stay time and package card click frequency from the user package adaptation type, and obtain the preliminary package recommendation list by matching the flow card package preferences of similar user groups.

[0041] The stay time of the member page and the click frequency of the package card are extracted from the user package adaptation type, the cumulative stay seconds and the click times of the user on each package detail page are recorded, and the sequence and the jump path of the user accessing the package page are collected, to obtain a user page interaction behavior data set. According to the user page interaction behavior data set, the attention index of each package is calculated, the stay time is multiplied by the click frequency to obtain the package attention value, by comparing the interaction behavior characteristics of the user groups with the same package adaptation type, the user set with similar behaviors is identified and the historical package purchase records of the user set are extracted. Based on the historical package purchase records, the frequency of each package being selected is counted as a group preference value, the package attention value is multiplied by 0.6, and the group preference value is multiplied by 0.4 to calculate a comprehensive recommendation score, and the preliminary package recommendation list is generated in descending order of the comprehensive recommendation score.

[0042] In an embodiment, when extracting page interaction data from the user package adaptation type, the stay duration records the complete time from the user entering the package detail page to leaving, accurate to the second level. The click frequency of the package card counts the number of times the user clicks on each package card on the member page, including viewing details, expanding explanations, and comparing functions. The access sequence records the time sequence of the user browsing different packages, and the jump path tracks the complete trajectory of the user from the package list page to the detail page to the purchase page.

[0043] It should be noted that the package attention index reflects the degree of interest of the user in a specific package.

[0044] Specifically, if the user stays on the 30GB monthly package detail page for 120 seconds and clicks 3 times, the attention value of the package is 360, while the 10GB basic package stays for 30 seconds and clicks 1 time, the attention value is only 30, and the difference between the two clearly reflects the user preference.

[0045] Preferably, the process of identifying behaviorally similar users is based on multi-dimensional feature comparison.

[0046] In a possible implementation, users with the same package adaptation type are extracted as a candidate set, and their page stay duration distribution, click frequency pattern, and access path similarity are compared. When the stay duration difference of two users is within 20%, the click frequency pattern correlation coefficient is more than 0.7, and more than 60% of the same nodes are in the access path, it is determined that the two users are behaviorally similar. Through this way, the similar user set identified has a strong reference value in historical package purchase records, because similar browsing behavior often indicates similar purchase decisions.

[0047] Illustratively, the statistics of historical package purchase records cover transaction data within the last 30 days. If there are 100 users in the similar user set, 60 of whom have purchased the 20GB peak speed-up package, 30 of whom have purchased the 30GB monthly package, and 10 of whom have purchased the 10GB basic package, the group preference values are 0.6, 0.3, and 0.1 respectively. Further, the calculation of the comprehensive recommendation score adopts a weighted summation method, with the package attention value weight 0.6 reflecting the importance of individual user behavior, and the group preference value weight 0.4 reflecting the reference value of group wisdom.

[0048] For example, the attention value of a certain package is 300, and the group preference value is 0.4, so the comprehensive recommendation score is 300x0.6+0.4x0.4x1000, where 1000 is the normalization coefficient.

[0049] It can be understood that the preliminary package recommendation list is arranged in descending order of the comprehensive recommendation score, and the package with a high score is displayed first, which improves the efficiency of the user finding a suitable package.

[0050] S104, analyze the user page jump behavior according to the member page browsing path, identify the current user behavior mode, if the user behavior mode is high-frequency interaction, evaluate the matching degree of real-time interaction behavior and user package adaptation type, prioritize the preliminary package recommendation list, and determine the optimized package combination scheme.

[0051] According to the member page browsing path, the jump sequence of the user between each page node is extracted, the page switching frequency in unit time is divided by the access time to obtain the interaction frequency index, the repeated access times of the package card are counted to obtain the user page interaction feature set. The behavior mode recognition is performed on the user page interaction feature set, if the interaction frequency index is more than 3 times per minute and the repeated access times of a single package are more than 5 times, it is determined that the high-frequency interaction mode is determined, the package sequence and the stay duration of the user access under the high-frequency interaction mode are recorded, and the current behavior mode identifier and its associated data are obtained. Based on the current behavior mode identifier and its associated data, the matching degree is evaluated, the coincidence number of the user real-time access package type and its package adaptation type is counted, the coincidence number is divided by the total access package number to obtain the consistency score, and the consistency score is multiplied by 100 as the priority value of the corresponding package in the preliminary package recommendation list to obtain the package priority sequence. According to the package priority sequence, the preliminary package recommendation list is arranged in descending order, the package with priority value more than 60 is selected as the main promotion package, and the package with priority value between 40 and 60 is selected as the alternative package, and the optimized package combination scheme containing the main promotion and alternative packages is formed.

[0052] In an embodiment, the extraction of the member page browsing path is realized by recording the jump behavior of the user between different page nodes.

[0053] Specifically, when the user clicks into the 20GB package detail page from the package list page, and then returns to the list page and enters the 30GB package detail page, this jump sequence is recorded as "list→20GB detail→list→30GB detail". The calculation of the interaction frequency index is based on the page switching behavior of the user in unit time, if the user completes 15 page switching in 5 minutes, the interaction frequency index is 3 times per minute. The repeated access times of the package card count the repeated viewing behavior of the user to the same package detail page, when the user returns to view the details, cost explanation and preferential terms of the same package multiple times in a session, each access is counted in the repeated access times. By integrating the jump sequence, the interaction frequency index and the repeated access times, a complete user page interaction feature set is formed, which contains the timing information, frequency characteristics and focus of the user browsing behavior.

[0054] It should be noted that the core of the behavior pattern recognition is to determine whether the user is in an active package selection state. When the interaction frequency index exceeds 3 times per minute, it indicates that the user is quickly browsing and comparing different packages; when the number of repeated accesses to a single package exceeds 5 times, it indicates that the user has strong interest in the package but may hesitate to make a decision.

[0055] Preferably, the determination of the high-frequency interaction mode adopts a double condition verification mechanism.

[0056] In one possible implementation, not only the numerical conditions of the interaction frequency and the number of repeated accesses are met, but also it is verified whether the behaviors occur within a continuous time window. If the user's high-frequency interaction behaviors are completed within 10 minutes, it is determined as an effective high-frequency interaction mode; if the behaviors are scattered in a longer period of time, even if the numerical conditions are met, it is not determined as a high-frequency interaction. The current behavior pattern identifier contains a pattern type label and timestamp information, and the associated data records the complete package sequence accessed by the user in this mode, including the access order, the stay duration and the page depth of each package.

[0057] Exemplarily, the matching degree evaluation process realizes accurate judgment by comparing the user's real-time behavior with the preset package adaptation type. Assuming that the user's package adaptation type is large-flow consumption type, the corresponding recommended packages include 30GB monthly package, 50GB quarterly package and unlimited annual package. In real-time interaction, the user accesses 5 packages in turn, including 30GB monthly package, 20GB monthly package, 50GB quarterly package, 30GB monthly package and 10GB basic package, of which 30GB monthly package and 50GB quarterly package belong to the recommended range of large-flow consumption type, and the number of coincidences is 2, and the consistency score is calculated as 2 divided by 5, equal to 0.4. Multiply the consistency score by 100 to get the priority value 40, which is assigned to 30GB monthly package and 50GB quarterly package. For the 30GB monthly package repeatedly accessed by the user, the priority value is additionally increased by the number of repeated accesses multiplied by 5 on the original basis, if the package is accessed 2 times, the final priority value is 40 plus 10, equal to 50. Further, the formation of the package priority sequence not only considers the priority value of a single package, but also introduces the relevance analysis between packages.

[0058] In an embodiment, when the priority values of multiple packages are similar, the comparison relationship between the packages is determined by analyzing the user's access path. If the user views the 50GB quarterly package immediately after viewing the 30GB monthly package, it indicates that the user is comparing the capacity and duration, and the two packages are marked as a comparison group and kept adjacent positions in the final sorting. The optimized package combination scheme adopts a hierarchical display strategy. The main promotion package is the package with a priority value exceeding 60, which is highly matched with the user's demand and is displayed in a large card form on the member page, including complete package description and one-click purchase button. The priority value of the alternative package is between 40 and 60, indicating that it is partially matched with the user's demand and is displayed in a small card form below the main promotion package, providing basic information and a view details entry. The package with a priority value less than 40 is classified as other optional packages and is displayed in a folded form at the bottom of the page.

[0059] For example, after behavior analysis, the priority value of the 30GB monthly package is 75, the priority value of the 20GB peak speed package is 65, the priority value of the 50GB quarterly package is 45, and the priority value of the 10GB basic package is 25. The package combination scheme displays the 30GB monthly package and the 20GB peak speed package as the main promotion package, displays the 50GB quarterly package as the alternative package in the secondary position, and folds the 10GB basic package in the other options. This dynamic optimization mechanism can adjust the recommendation strategy according to the user's real-time behavior, quickly respond and optimize the display order when the user shows clear package preferences, reduce the user's selection cost, and improve the package purchase conversion rate.

[0060] S105, adjust the display order of the traffic card package on the member page according to the optimized package combination scheme, top the package scheme according to the click frequency, and obtain a personalized package streaming scheme.

[0061] According to the optimized package combination scheme, the recommended priority value of each package is extracted, the number of clicks of each package is counted from the user's historical behavior data, and the number of clicks is normalized and added to the recommended priority value to obtain a comprehensive display weight. The display order of the packages is determined from high to low according to the comprehensive display weight. The page display parameters are configured according to the display order of the packages, the top three packages in the comprehensive display weight are set to top display, a large size card and a prominent visual effect are allocated, the packages with weight ranking from four to eight use a medium size card, and the remaining packages use a compact list form, obtaining a hierarchical package display layout configuration. Based on the package display layout configuration, a display scheme is generated, the display area of the top package is set to 1.5 times the standard card and a dynamic effect is configured, the medium weight package maintains the standard card size, and the low weight package only displays basic information, forming a personalized package streaming scheme including position sorting, size allocation and visual intensity.

[0062] In an embodiment, the calculation of the comprehensive display weight is realized by using a weighted fusion method to achieve dynamic adjustment.

[0063] Specifically, the recommended priority value extracted from the optimized package combination scheme ranges from 0 to 100, reflecting the matching degree of the package and the user's demand. The normalization processing of the click times maps the historical click data to the interval of 0 to 1, and the normalized value of the package with the most clicks is 1, and the package that is not clicked is 0. The comprehensive display weight is calculated by adding the recommended priority value and multiplying the normalized click times by 20, so that the contribution of historical behavior to the maximum weight is controlled within a reasonable range. The determination of the package display order follows the weight priority principle, and the package with a high comprehensive display weight obtains a better display position, which improves the probability of attention acquisition of the user. The hierarchical display layout configuration is dynamically adjusted according to the distribution characteristics of the comprehensive display weight.

[0064] In a possible implementation, when the top three package weight values all exceed 80, a parallel top display is used; if only one package weight exceeds 80, it is displayed alone, and the remaining packages are arranged in descending order of weight. The large-size card provided for the top package contains the package name, traffic capacity, validity period, main benefits and time-limited preferential information, and occupies more than 90% of the page width. The medium-size card displays the basic information and price of the package, and occupies 70% of the page width. The compact list form only displays the package name and price, and is displayed in a single row, which is convenient for quick browsing.

[0065] Exemplarily, the dynamic effects of the top package include gradient background, micro-animation and highlight border. When the user slides the page, the top package remains in a fixed position for 3 seconds and then scrolls with the page, ensuring sufficient exposure. The display area is set to 1.5 times the size of the standard card, which means an increase of 50% in height and no change in width, leaving space for additional promotional information and user reviews. Further, the personalized package streaming scheme is dynamically updated according to the user's real-time behavior.

[0066] For example, when the user stays on a certain package card for more than 5 seconds, the display weight of that package is temporarily increased by 10 points, and it is automatically adjusted to a more forward position at the next page refresh.

[0067] It can be understood that this hierarchical display mechanism guides the user's attention through visual levels, improves the conversion probability of high-matching-degree packages, and at the same time preserves the display opportunities of other packages, achieving a balance between recommendation efficiency and selection diversity.

[0068] The recommendation weight and user matching degree of each package are obtained from the optimized package combination scheme, the display positions of the packages on the member page are arranged according to the recommendation weight and user matching degree, the package with the highest matching degree is preferentially displayed in the user's visual focus area, the visual prominence and display area of the package card are adjusted according to the user's historical click behavior, the order and visual weight of the package display are dynamically adjusted, the user is guided to pay attention to the traffic card scheme that best meets his needs, and a personalized page display layout is formed.

[0069] The recommendation weight value and user matching degree score of each package are obtained from the optimized package combination scheme, the order of the packages is arranged from high to low according to the matching degree score, the click number and page stay duration of each package in the user's historical click behavior are extracted, and a package display basic data set is obtained. Visual parameter calculation is performed on the package display basic data set, the vertical position of the package is determined according to the matching degree score, the package with the highest matching degree is positioned in the top one-third area of the page as the visual focus area, the visual prominence coefficient is calculated by multiplying the click number by 0.7 and adding the stay duration in seconds by 0.3, and a package visual configuration parameter set is obtained. The display attribute is adjusted based on the package visual configuration parameter set, if the visual prominence coefficient exceeds a preset threshold, the display area of the package card is increased to 1.2 to 1.5 times of the standard area, the color saturation is increased by 20% and the border is widened by 2 pixels, the left and right order of the package in the same row is adjusted according to the recommendation weight value, and an adjusted display attribute configuration is obtained. The page layout is constructed according to the adjusted display attribute configuration, the page is horizontally divided by using a 12-column grid, the number of grids occupied by the package is allocated according to the display area, the packages with high visual prominence coefficients are added with gradient backgrounds and entrance animations, and a personalized page display layout containing position, size and style is formed.

[0070] In an embodiment, the construction of the package display basic data set involves the integration processing of multi-source data.

[0071] Specifically, the recommendation weight value is derived from the output of the pre-procedure recommendation algorithm, with a value range of 0 to 100, where 80 or above represents a strong recommendation, 60 to 80 represents a moderate recommendation, and 60 or below represents an alternative recommendation. The user matching score is calculated by comparing the user portrait with the package feature vector, using the cosine similarity algorithm to match the user's traffic demand, price sensitivity, usage time distribution, and other multi-dimensional characteristics with the package's capacity, cost, and preferential time attributes. The historical click behavior data includes the user's click count on each package in the past 30 days, the page dwell time after each click, whether to view the detail page, whether to add to the comparison list, and other behavior sequences. These data are processed with time decay, with recent behavior given a higher weight, forming a comprehensive package display base dataset. The determination of the visual focus area is based on the golden section principle of eye tracking research, with the top third of the page being the natural focus of the user's gaze, with an information acquisition rate of 75% or more.

[0072] Preferably, the calculation of the visual prominence coefficient uses a weighted fusion method, multiplying the click count by a weight coefficient of 0.7 and the dwell time in seconds by a weight coefficient of 0.3, and adding the two to obtain the initial visual prominence coefficient.

[0073] In one possible implementation, if a user clicks on a certain package 10 times with an average dwell time of 60 seconds, the visual prominence coefficient is calculated as 10 x 0.7 + 60 x 0.3. To avoid the influence of extreme values, the calculation result is normalized, mapping the visual prominence coefficients of all packages to the standard interval of 0 to 100. When the visual prominence coefficient exceeds 70, the package is determined to be a high-attention package for the user, exceeding 50 is a moderate-attention package, and below 50 is a low-attention package. This hierarchical mechanism ensures that packages with different attention levels receive differentiated visual presentation, with high-attention packages attracting user attention through enhanced visual effects, and low-attention packages displayed in a simple form to avoid information overload.

[0074] Illustratively, the adjustment of the display attributes involves the coordinated change of multiple visual dimensions. The adjustment range of the display area is set between 1.2 to 1.5 times the standard area, with 1.2 times suitable for moderate-attention packages and 1.5 times suitable for high-attention packages. The color saturation is increased by adjusting the S component of the HSL color space, increasing the original saturation value by 20%, enhancing the visual impact of the package card. The border width is increased from the standard 1 pixel to 3 pixels, with a gradient color border transitioning from the package theme color to white, forming a soft visual boundary. Further, the ranking of the recommendation weight values within the same row follows the principle of decreasing from left to right, conforming to the user's reading habits. When the recommendation weight values of multiple packages are similar, secondary sorting factors are introduced, including the price-performance ratio index of the package and the probability of the user's historical purchase of similar packages.

[0075] In an embodiment, the 12-column grid system evenly divides the page horizontally, with each column occupying 8.33% of the total page width. High-attention packages occupy 6-8 columns, ensuring sufficient display space; medium-attention packages occupy 4-5 columns, maintaining moderate information density; and low-attention packages occupy 2-3 columns, presented in a compact form. The responsive design of the grid system automatically adjusts column widths on different screen sizes, with mobile devices compressing 12 columns to 4 columns and tablets adjusting to 8 columns, ensuring consistent experience across devices.

[0076] For example, the gradient background uses a linear gradient that transitions from 10% transparency of the package theme color to full transparency, with a gradient angle of 135 degrees, creating a visual flow from the top left to the bottom right.

[0077] It can be understood that the design of the entrance animation takes into account the cognitive load of the user, using a progressive loading strategy. The animation duration of high-attention packages is 0.5 seconds, including a fade-in effect and a slight upward movement of 20 pixels. The animation time of medium-attention packages is shortened to 0.3 seconds, containing only a fade-in effect. Low-attention packages have no animation effect and are displayed directly. The timing of the animation uses staggered triggering, with a delay of 0.1 seconds between each package, creating a wave-like visual rhythm.

[0078] Specifically, the form of the personalized page display layout contains three levels: the visual focus layer displays 1-3 high-matching packages, occupying the golden position at the top of the page; the transition layer displays 4-6 medium-matching packages, providing supplementary choices; and the background layer displays the remaining packages in a list form, maintaining the integrity of the information. This layered layout guides the user's decision-making path through visual levels, improving the efficiency and accuracy of package selection.

[0079] S106, from the personalized package streaming scheme, obtain the user package click conversion rate, through continuous integration of newly collected viewing time, identify user preference change trend to determine recommendation weight configuration.

[0080] The number of clicks and the number of purchases of each package are extracted from the personalized package streaming scheme, the number of purchases is divided by the number of clicks to obtain the click conversion rate of each package, the latest one-week short drama viewing time data of the user is obtained, and the new data is appended to the end of the historical viewing record to form a time series data set. The preference change detection is performed on the time series data set, a 7-day sliding window size is set, the difference between the daily average viewing time in the current window and the previous window is calculated, if the absolute value of the difference exceeds 30 minutes, it is marked as a preference change point, the number of preference change points in 30 days is counted, when the number exceeds 3 and the difference is positive, it is determined as an upward trend, and when the difference is negative, it is determined as a downward trend. Based on the preference change trend and the click conversion rate, the recommendation weight is adjusted, if it is an upward trend, the recommendation weight of the large flow package is multiplied by an enhancement coefficient of 1.2, if it is a downward trend, the recommendation weight of the light package is multiplied by an enhancement coefficient of 1.3, and the recommendation weight of the package with a click conversion rate lower than 0.1 is multiplied by a decay coefficient of 0.8, to determine the updated recommendation weight configuration.

[0081] In an embodiment, the calculation of the click conversion rate is based on the actual purchase behavior of the user.

[0082] Specifically, the total number of clicks of each package during the display period is extracted from the execution log of the personalized package streaming scheme, and the purchase record of the corresponding package is queried from the order database. If a 30GB monthly package is clicked 50 times in a week, and 5 purchases are generated, the click conversion rate of the package is 0.1. The conversion rate directly reflects the effectiveness of the package recommendation, and becomes an important basis for subsequent weight adjustment. The construction of the time series data set adopts an incremental updating method, and the historical viewing records of the user in the past 60 days are retained, the viewing time data added every day is appended to the end of the sequence, and a continuous time series is formed. In the implementation process of the sliding window method, the selection of the 7-day window size is based on the periodic characteristics of user behavior.

[0083] In a possible implementation, the first window covers data of the 1st to 7th day, the daily average viewing time of these 7 days is calculated as 120 minutes; the second window covers the 2nd to 8th day, the daily average viewing time is 135 minutes, and the difference between the two windows is 15 minutes. When the sliding step is set to 1 day, a new difference data point is generated every day. The threshold of 30 minutes is set considering the normal fluctuation range of user viewing habits, and exceeding this threshold indicates that the user's viewing behavior has changed substantially.

[0084] Exemplarily, the judgment of the preference change trend follows the continuity principle. When there are 4 preference change points in a 30-day observation period, 3 of which have a positive difference of 40 minutes and 1 has a positive difference of 25 minutes, it is determined that there is an obvious upward trend, indicating that the user's demand for short video content is increasing. On the contrary, if most of the differences are negative, it is determined that there is a downward trend. Further, the adjustment of the recommendation weight adopts a differentiation coefficient strategy. The upward trend corresponds to an enhancement coefficient of 1.2, which increases the recommendation weight of the large-flow package from the original 80 to 96, improving the priority of such packages in the recommendation list. The 1.3 enhancement coefficient triggered by the downward trend strengthens the recommendation of the light package, meeting the user's demand for reducing traffic consumption. Packages with a click-through rate lower than 0.1 are considered to have insufficient appeal, and a decay coefficient of 0.8 reduces their display frequency, freeing up display space for high-conversion packages.

[0085] It can be understood that this dynamic adjustment mechanism realizes the real-time synchronization of the recommendation strategy and the user behavior, improving the accuracy and conversion efficiency of the package recommendation.

[0086] S107, for the recommendation weight configuration, a feedback mechanism is used to reprocess the behavior pattern recognition. If the matching degree between the dynamic demand change and the real-time interaction behavior is improved, the final traffic card package recommendation output is obtained, including the package type, traffic capacity, price level, and recommendation reason.

[0087] For the recommendation weight configuration, a feedback mechanism is used to obtain the click and purchase behavior data of the user on the recommended package, compare the actual purchased package with the recommended package, calculate the recommendation accuracy as a deviation value, update the threshold parameter of the behavior pattern recognition according to the deviation value multiplied by the adjustment coefficient, re-execute the user behavior pattern recognition, and obtain the adjusted behavior pattern identification. Based on the adjusted behavior pattern identification, the dynamic demand features are extracted, the change rate of daily traffic consumption is calculated from the user's traffic usage records in the past 7 days, the number of times the user views different package detail pages is counted, the user's current page browsing and click behavior sequence is obtained, and the cosine similarity between the dynamic demand feature vector and the behavior sequence vector is calculated as the matching degree value. If the matching degree value exceeds a preset threshold of 0.7, the package information corresponding to the user behavior pattern is extracted from the package library, including the package type, traffic capacity value, and price level, the recommendation reason description is generated according to the user's viewing time growth trend and traffic usage features, and if the matching degree value does not exceed the threshold, the user behavior data is re-collected. The package information and the recommendation reason description are integrated, and the final traffic card package recommendation result is output in a structured format according to the package type, traffic capacity, price level, and recommendation reason.

[0088] In an embodiment, the implementation of the feedback mechanism adopts the principle of closed-loop control to optimize the recommendation strategy by continuously monitoring the user's response to the recommended results. After the user receives the package recommendation, the feedback of his behavior within 24 hours is recorded, including whether to click to view details, whether to join the comparison list, whether to finally purchase, and whether to purchase the recommended package or other packages. The calculation of recommendation accuracy is based on a multi-level scoring mechanism: if the user purchases the first ranked recommended package, the accuracy is recorded as 1.0; if the user purchases the second ranked recommended package, the accuracy is recorded as 0.8; if the user purchases the third ranked recommended package, the accuracy is recorded as 0.6; and if the user purchases a non-recommended package, the accuracy is recorded as 0. This gradient scoring method can more accurately reflect the quality of the recommendation, rather than a simple binary judgment. The update of the behavior pattern recognition parameter adopts an adaptive adjustment strategy. The adjustment coefficient is dynamically set according to different intervals of the recommendation accuracy: when the accuracy is less than 0.3, the adjustment coefficient is set to 1.5, indicating that it needs to be corrected significantly; when the accuracy is between 0.3 and 0.7, the adjustment coefficient is 1.2, for moderate adjustment; and when the accuracy is higher than 0.7, the adjustment coefficient is 1.05, for fine tuning only.

[0089] Preferably, the extraction of dynamic demand features covers multiple dimensions of user behavior changes.

[0090] In a possible implementation, the calculation of traffic consumption change rate adopts the moving average method, compares the daily average traffic consumption of the past 7 days with that of the previous 7 days, and obtains the percentage of increase or decrease. The change in package viewing frequency is identified by counting the number of times the user views the package detail page at different times, and if the number of views at night increases from an average of 2 to 5, it indicates that the user's attention to packages has increased. The construction of the behavior sequence records the user's page access path, including the complete trajectory from entering the member page from the home page, browsing the package list, clicking on a specific package to view details, and returning to the list to continue browsing. These multi-dimensional features together constitute a panoramic portrait of the user's current demand state, providing a rich data foundation for subsequent matching degree calculation.

[0091] Illustratively, the calculation of cosine similarity maps the dynamic demand feature vector and the behavior sequence vector to the same vector space. The dynamic demand feature vector contains dimensions such as traffic growth rate, peak period usage ratio, and large capacity package viewing frequency, each of which is assigned a different weight according to its importance. The behavior sequence vector converts the user's operation sequence into a numerical representation, such as encoding the sequence "view 30GB package → view 50GB package → return to 30GB package" as a preference intensity value for large traffic packages. The setting of the matching degree threshold of 0.7 is based on statistical analysis of historical data, and when the matching degree exceeds this threshold, the conversion rate of the recommended package purchase reaches more than 35%, which has high commercial value.

[0092] In an embodiment, the generation of the recommendation reason adopts a combination of templates and dynamic content. The basic template contains guiding language such as "based on your viewing habits", "considering your traffic demand", etc., and the dynamic content is filled according to the specific characteristics of the user.

[0093] For example, if the user's viewing duration shows an upward trend and often watches high-definition content, the recommendation reason is described as "your recent short drama viewing duration has increased by 40%, and the high-definition viewing ratio has reached 70%. Recommend this high-traffic package to meet your viewing needs". When the matching degree value does not exceed the threshold value, the data re-collection mechanism is triggered instead of forced recommendation, which avoids the negative impact of low-quality recommendation on user experience.

[0094] Specifically, the structured output of the final recommendation result adopts a standardized format, the package type field contains the classification identification of monthly card, quarterly card and annual card; the traffic capacity is represented in integer form, with GB as the unit; the price level is divided into three levels of economy, standard and luxury; and the recommendation reason is limited to within 100 characters to ensure the simplicity and readability of the information. This structured output format facilitates front-end display and user understanding, and improves the efficiency of conveying the recommendation information.

[0095] For example, a complete recommendation output contains the complete information of "30GB monthly card, 30, standard type, based on your daily viewing of 3 hours of short dramas and preference for high-definition quality, this package can meet your monthly traffic demand", realizing the closed loop of personalized recommendation.

[0096] The application provides a cloud big data real-time streaming system based on intelligent analysis, mainly comprising: A behavior data collection module is used to collect real-time interactive behaviors of users, including short drama viewing duration and member page interaction records, and to obtain a standardized behavior data set through format processing; A user preference identification module is used to identify viewing duration and quality preference through the standardized behavior data set, to determine the user traffic consumption type according to the viewing duration and quality preference, and to obtain the user package adaptation type; A preliminary recommendation generation module is used to extract member page stay duration and package card click frequency statistics from the user package adaptation type, to obtain a member page browsing path, to match the traffic card package preference of similar user groups, and to obtain a preliminary package recommendation list; A behavior pattern analysis module is used to analyze user page jump behaviors according to the member page browsing path, to identify the current user behavior pattern, to evaluate the matching degree of real-time interactive behaviors and the user package adaptation type if the user behavior pattern is high-frequency interaction, to prioritize the preliminary package recommendation list, and to determine an optimized package combination scheme; A personalized flow injection module is configured to adjust the display order of the flow card package on the member page according to the optimized package combination scheme, to top the package scheme according to the click frequency, and to obtain a personalized package flow injection scheme; A weight configuration updating module is configured to obtain a user package click conversion rate from the personalized package flow injection scheme, to identify a user preference change trend by continuously integrating newly collected viewing time, and to determine a recommendation weight configuration. A dynamic recommendation output module is configured to reprocess behavior pattern recognition by using a feedback mechanism for the recommendation weight configuration, to obtain a final flow card package recommendation output including a package type, a flow capacity, a price level, and a recommendation reason if the dynamic demand change and the real-time interaction behavior matching degree are improved.

[0097] The above is only a specific implementation of the present specification, and those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, module and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here. It should be understood that the protection scope of the present specification is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed in the present specification, and these modifications or replacements should be covered within the protection scope of the present specification.

Claims

1. A cloud big data real-time streaming method based on intelligent analysis, characterized in that, The method comprises: By collecting real-time interaction behavior of users, standardized behavior data sets are obtained by format processing, wherein the standardized behavior data sets contain short play viewing duration and member page interaction records; by the standardized behavior data sets, user viewing duration and quality preference are identified, and user package adaptation types are determined according to the viewing duration and quality preference; from the user package adaptation types, member page stay duration and package card click frequency are extracted, member page browsing paths are obtained, traffic card package preferences of similar user groups are matched, and a preliminary package recommendation list is generated; according to the member page browsing paths, user page jump behaviors are analyzed, user behavior patterns are identified, and the matching degree of real-time interaction behavior and the user package adaptation types is evaluated according to the user behavior patterns, the preliminary package recommendation list is sorted, and an optimized package combination scheme is determined; according to the optimized package combination scheme, the display order of traffic card packages in the member page is adjusted, the top package is determined according to the click frequency, and a personalized package streaming scheme is generated; from the personalized package streaming scheme, user package click conversion rates are extracted, user preference trends are identified in combination with newly collected viewing duration, and recommendation weight configurations are determined; according to the recommendation weight configurations, behavior pattern recognition is reprocessed, and the final traffic card package recommendation results are output according to the matching degree of dynamic demand changes and real-time interaction behaviors, including package types, traffic capacities and price levels.

2. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, The standardized behavior data sets are obtained by collecting real-time interaction behavior of users, and the method comprises: Real-time operation records and member page interaction records of users in a short play player are obtained, short play viewing duration, pause times, quality switching records, package card click coordinates and page stay duration are extracted, and original behavior sequences are formed according to timestamp alignment; abnormal records are removed from the original behavior sequences, data formats are converted, and the standardized behavior data sets are generated by normalizing the value range to the interval of 0 to 1.

3. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, The user viewing duration and quality preference are identified from the standardized behavior data sets, and the user package adaptation types are determined according to the viewing duration and quality preference, which comprises: From the standardized behavior data sets, viewing duration sequences and quality selection records are extracted, the ratio of daily viewing duration cumulative value to preset duration is calculated as a viewing density index, the proportion of standard definition, high definition and ultra definition selection times is counted to form a quality preference distribution, and a user viewing behavior feature vector is formed by combination; according to the user viewing behavior feature vector, the traffic consumption in unit time is calculated by combining the quality selection frequency and the code rate standard, the large traffic consumption type, the high-speed traffic consumption type or the light traffic consumption type is determined according to the viewing density index and the quality preference distribution, the corresponding package capacity is matched, and the user package adaptation types are generated.

4. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, The member page stay duration and package card click frequency are extracted from the user package adaptation types, the member page browsing paths are obtained, the traffic card package preferences of similar user groups are matched, and the preliminary package recommendation list is generated, which comprises: Extract the member page stay time and package card click frequency from the user package adaptation type, record the page jump path, and generate a user page interaction behavior dataset; calculate the package attention value according to the user page interaction behavior dataset, combine the historical package purchase records of the similar user group to calculate the group preference value, calculate the comprehensive recommendation score, and sort to generate the preliminary package recommendation list.

5. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, According to the member page browsing path, analyze the user page jump behavior, identify the user behavior mode, evaluate the matching degree of real-time interaction behavior and user package adaptation type according to the user behavior mode, sort the preliminary package recommendation list, and determine the optimized package combination scheme, including: According to the member page browsing path, extract the page jump sequence, calculate the interaction frequency index and the number of repeated visits to the package card, and generate a user page interaction feature set; determine the high-frequency interaction mode according to the user page interaction feature set, record the package access sequence and stay time; calculate the coincidence ratio of real-time access package type and the user package adaptation type as the consistency score, assign a package priority value based on the consistency score, sort the preliminary package recommendation list, and generate the optimized package combination scheme containing the main push and alternative packages.

6. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, According to the optimized package combination scheme, adjust the display order of the traffic card package in the member page, top the package according to the click frequency, and generate a personalized package streaming scheme, including: Extract the recommendation priority value from the optimized package combination scheme, calculate the comprehensive display weight by combining the user historical click frequency, and determine the package display order; configure the page display parameters according to the comprehensive display weight, set the display area and visual effect of the top package, medium weight package and low weight package, and generate the personalized package streaming scheme containing position sorting and style.

7. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, According to the optimized package combination scheme, adjust the display order of the traffic card package in the member page, top the package according to the click frequency, and generate a personalized package streaming scheme, including: Extract the recommendation weight value and user matching degree score from the optimized package combination scheme, calculate the visual prominence coefficient by combining the user historical click frequency and stay time, and generate a package visual configuration parameter set; adjust the package display area, color saturation and border style according to the package visual configuration parameter set, allocate the page grid number, add dynamic effects, and generate a personalized page display layout containing position, size and style.

8. The cloud big data real-time streaming method based on intelligent analysis according to claim 1, characterized in that, Extract the user package click conversion rate from the personalized package streaming scheme, identify the user preference change trend by combining the newly collected viewing time, and determine the recommendation weight configuration, including: Extract the click frequency and purchase frequency from the personalized package streaming scheme, calculate the click conversion rate, combine the latest viewing time data to form a time series dataset; detect the preference change point according to the time series dataset, count the change trend, adjust the recommendation weight of the high-traffic package or lightweight package, and generate the recommendation weight configuration. 9.The smart analysis based cloud big data real-time streaming method of claim 1, wherein, The behavior pattern recognition is reprocessed according to the recommended weight configuration, a matching degree of dynamic demand change and real-time interaction behavior is output, and a final traffic card package recommendation result is output, including a package type, a traffic capacity and a price level. According to the recommended weight configuration, a behavior pattern recognition threshold is updated in combination with user click and purchase behavior data, a behavior pattern identifier is regenerated, dynamic demand characteristics are extracted according to the behavior pattern identifier, a traffic consumption change rate and a package detail page access frequency are calculated, and a matching degree value is determined, a package type, a traffic capacity value and a price level are extracted according to the matching degree value, and a structured final traffic card package recommendation result is generated.

10. A cloud big data real-time streaming system based on intelligent analysis, characterized in that, The system comprises: A behavior data acquisition module is configured to acquire real-time interaction behavior of a user, including short drama viewing time and member page interaction records, and to obtain a standardized behavior data set through format processing; A user preference identification module is configured to identify viewing time and picture quality preferences through the standardized behavior data set, to determine a user traffic consumption type according to the viewing time and the picture quality preferences, and to obtain a user package adaptation type; A preliminary recommendation generation module is configured to extract member page stay time and package card click frequency statistics from the user package adaptation type, to obtain a member page browsing path, to match traffic card package preferences of similar user groups, and to obtain a preliminary package recommendation list; A behavior pattern analysis module is configured to analyze user page jump behavior according to the member page browsing path, to identify a current user behavior pattern, and to evaluate a matching degree of real-time interaction behavior and the user package adaptation type if the user behavior pattern is a high-frequency interaction, to prioritize the preliminary package recommendation list, and to determine an optimized package combination scheme; A personalized flow injection module is configured to adjust a display order of a traffic card package on a member page according to the optimized package combination scheme, to top a package scheme according to a click frequency, and to obtain a personalized package flow injection scheme; A weight configuration update module is configured to obtain a user package click conversion rate from the personalized package flow injection scheme, to identify a user preference change trend by continuously integrating newly acquired viewing time, and to determine a recommended weight configuration; A dynamic recommendation output module is configured to reprocess behavior pattern recognition by using a feedback mechanism for the recommended weight configuration, and to obtain a final traffic card package recommendation output, including a package type, a traffic capacity, a price level and a recommendation reason, if a matching degree of dynamic demand change and real-time interaction behavior is improved.

Images