A method of recording campus life by integrating large models

Abstract

The present application provides a campus life recording method integrating a large model, including: determining the course availability information of each time period based on the course schedule data of the whole school, and judging the course availability and the demand for the use of activity venues during the activities in combination with the school activity data; obtaining the user's personal course schedule data, extracting the user's free time data, and at the same time, analyzing the user's exercise characteristics from the user's historical exercise data, including exercise type, exercise intensity and exercise habits, and establishing a user exercise characteristic table; regularly collecting user feedback data, including the user's satisfaction with the optimal exercise time, the difference between the actual exercise time and the optimal exercise time, and updating the user exercise characteristic table according to the feedback data.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a campus life recording method integrating a large model. Background Art

[0002] In the process of developing prediction models and personalized recommendation algorithms, we are faced with the technical contradiction between data integration and algorithm optimization. First, the school's curriculum data and key dates in the school calendar are scattered in different systems, and the data formats are inconsistent, so data cleaning and standardization are required. In addition, different types of courses and the nature of activities have different effects on the use of venues. It is necessary to design feature engineering for different scenarios and extract effective feature representations. Users' exercise habits and time arrangements are different. Some users may pay more attention to the flexibility of exercise time, while others may care more about the comfort of the exercise environment. How to design a recommendation strategy that takes into account both personalization and global optimization requires comprehensive consideration of user feedback and system performance. In short, in the process of integrating campus data, establishing prediction models, and developing personalized recommendation algorithms, it is necessary to balance multiple factors such as data quality, model performance, and user experience, and continuously iterate and optimize to achieve intelligent sports venue management and personalized sports guidance services. Summary of the invention

[0003] The present invention provides a campus life recording method integrating a large model, which mainly includes:

[0004] By collecting the school's curriculum data, including the class time and location of each class, important dates in the school calendar, and the schedule of school activities, the curriculum data is integrated and standardized to build a school-wide curriculum and activity database;

[0005] Based on the school's curriculum data, determine the course availability information for each time period, and combine it with school activity data to determine the course availability and demand for activity venues during activity periods;

[0006] Set up a prediction model based on time series prediction or regression analysis, input the school's curriculum schedule, school activities and historical venue usage data, and output the predicted value of venue usage rate in each time period by training and optimizing the prediction model parameters, and identify the expected peak period;

[0007] Obtain the user's personal course schedule data, extract the user's free time data, and at the same time, analyze the user's exercise characteristics from the user's historical exercise data, including exercise type, exercise intensity and exercise habits, and establish a user exercise characteristic table;

[0008] Based on the user's sports feature table and the venue's peak hours, collaborative filtering or content-based recommendation models are used to generate personalized sports time recommendations for each user, including obtaining the user's best sports time based on the user's free time, sports features, and venue's peak hours;

[0009] Regularly collect user feedback data, including user satisfaction with the optimal exercise time, the difference between the actual exercise time and the optimal exercise time, and update the user exercise feature table based on the feedback data.

[0010] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0011] The present invention discloses a campus life recording method integrating a large model. The core of the method is to optimize the utilization efficiency of campus resources through refined management and scientific analysis, while improving the sports experience of students and faculty. Specific business scenario problems include high-load use of campus venues during peak periods and unreasonable allocation of students' personal sports time. The present invention constructs a school-wide course and activity database by integrating and standardizing the course schedule data of the whole school. This enables the present invention to accurately obtain the course idle information of each time period, and effectively judge the actual use demand of the activity venue during the course idle period and the activity period in combination with the school activity data. A prediction model based on time series prediction or regression analysis is set. By inputting the course schedule, school activities and historical venue usage data of the whole school, the present invention can predict the venue utilization rate of each time period and accurately identify the expected peak period. This provides a scientific basis for venue management, helps to optimize venue allocation and plan resource use in advance. According to the user's personal course schedule data and historical exercise data, a user exercise feature table is established. This feature table covers multiple dimensions such as the user's exercise type, exercise intensity and exercise habits. A collaborative filtering or content-based recommendation model is used to generate personalized exercise time recommendations for each user. The recommendation model will comprehensively consider the user's free time, exercise characteristics and venue peak hours to recommend the best exercise time for the user. A user feedback mechanism is set up to regularly collect users' satisfaction with the recommended exercise time and the difference between their actual exercise time and the recommended exercise time. These feedback data will be used to update the user's exercise feature table to ensure the continuous optimization and efficient operation of the recommendation system. The technical effect of the present invention is to effectively improve the utilization efficiency of campus venues and the personalization of user exercise experience. Through scientific management and precise analysis, the present invention not only optimizes campus resource allocation, but also enhances users' health promotion and quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 The present invention is a flow chart of a campus life recording method integrating a large model.

[0013] Figure 2It is a schematic diagram of a campus life recording method integrating a large model according to the present invention.

[0014] Figure 3 It is another schematic diagram of a campus life recording method integrating a large model according to the present invention. DETAILED DESCRIPTION

[0015] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0016] like Figure 1-3 In this embodiment, a campus life recording method integrating a large model may specifically include:

[0017] Step S101, by collecting the school's curriculum data, including the class time and location of each class, important dates in the school calendar, and the schedule of school activities, the curriculum data is integrated and standardized to build a school-wide curriculum and activity database.

[0018] Obtain the course schedule data of all classes in the school, structure and standardize the course schedule data to form a course schedule data set; analyze the school calendar, extract the exact dates of the start and end of the semester, the start and end time range of the exam week, and the holiday arrangements, and determine whether each day is a normal class day through time overlap and intersection calculations, and associate and match the determination results with the course schedule data set; extract the time intervals of various seasonal activities of the school, compare the activity time intervals with the course schedule data set, identify temporary courses that need to be adjusted or suspended during the activities, and mark the temporary courses; construct a data model that combines the school calendar and the course schedule based on the collected time, location, course, and class.

[0019] For example, through data collection and integration technology, detailed curriculum information of all classes in the school is obtained, including the specific class time, course name and corresponding classroom of each class, etc., and these raw data are structured and standardized to eliminate the differences in data formats from different sources and form a unified and standardized curriculum data set. Analyze the key time nodes in the school calendar, extract the exact dates of the start and end of the semester, identify the start and end time range of the exam week, and obtain the arrangements for statutory holidays and school-defined holidays. Convert these time periods into a unified time format (such as date range or timestamp), use data structures such as interval trees to store these time periods, and design efficient algorithms to calculate time overlap and intersection. Through the query operation of the interval tree, determine whether each day is a normal class day, and associate and match this information with the curriculum data. Sort out various seasonal activities in the school, such as sports games, extract the time intervals for the activities, compare the activity arrangements with the daily curriculum, identify the courses that need to be adjusted or suspended during the activities, and mark the curriculum for these special periods separately. Based on the collected multi-dimensional information such as time, location, course, class, etc., a data model combining the school calendar and the curriculum is constructed. Use relational databases (such as MySQL) to store information such as curriculum, classes, and classrooms, and establish relationships between them through foreign key associations. Create indexes on key fields (such as class number, classroom number, course number, etc.) to speed up data query. Through data association and indexing technology, quickly query and locate the curriculum schedule of any time point and class, support multi-condition combination and cross query, use SQL statements for multi-condition combination query, and use JOIN operations to associate and filter data of different dimensions to improve information retrieval efficiency. Use data visualization technology to intuitively present the curriculum schedule and school calendar date, develop an interactive calendar view, allow users to quickly browse the curriculum schedule of different time periods by clicking on the date, switching weeks, etc., and provide filtering functions by dimensions such as class and course, so that users can conduct personalized information queries. Through big data analysis technology, mine and analyze the integrated curriculum and activity database to discover the rules and characteristics of the curriculum schedule. Use clustering algorithms (such as K-means) to group courses and discover the distribution characteristics of different types of courses. Apply association rule mining algorithms (such as Apriori) to analyze the correlation between different courses and discover common course combination patterns. Use decision trees or random forest algorithms to predict the matching degree between course schedule and teaching staff, identify potential resource conflicts or unreasonable arrangements, and provide data support and decision-making reference for schools to optimize resource allocation and reasonably arrange teaching activities.Establish a data update mechanism, connect with the teaching system, class scheduling system, etc., synchronize the changes of the class schedule in real time, and update the newly added school activities, temporary class adjustments and other information to the database in time to ensure that the integrated course and activity data remain accurate and timely at all times, and provide reliable query services for teachers and students. In the data collection and integration stage, crawler technology, such as Python's Scrapy framework, can be used to crawl the teaching system website and obtain the HTML source code of the class schedule. Use regular expressions or parsing libraries such as BeautifulSoup to parse the source code, extract the class time, course name, classroom and other information of each class, and convert it into structured JSON or CSV format data. For data from different sources, you can convert it into a unified format by writing data cleaning and conversion scripts, such as converting the time string "2023-05-0108:00:00" into the standard ISO8601 format "2023-05-01T08:00:00Z". When dealing with school calendars, time periods such as semester start and end dates, exam weeks, and holidays can be represented as tuples of start and end timestamps, such as (1621123200, 1621209600) for the time range from May 16 to May 17, 2021. Use interval tree algorithms, such as Python's intervaltree library, to insert these time periods into the interval tree. Given a date timestamp, such as 1621180800 for 12 noon on May 17, 2021, the query operation of the interval tree can determine whether the timestamp falls within a certain time period within the time complexity of O(logn), thereby determining whether it is a normal school day. For the storage of class schedule data, you can use the MySQL relational database to design tables such as the course schedule (id, course_id, class_id, classroom_id, start_time, end_time), class (id, name, grade), and classroom (id, name, capacity), and establish relationships between them through foreign key associations. Create a joint index on the start_time and end_time fields of the course table, and create a single-column index on the class_id and classroom_id fields to speed up data query. Use SQL statements for multi-condition combination queries, such as "SELECT * FROM schedule WHERE class_id = 101 AND start_time > = '2023-05-01 08:00:00' AND end_time < = '2023-05-01 12:00:00'", and associate the class and classroom tables through JOIN operations to implement data filtering by dimensions such as class and classroom. In the data analysis and mining stage, the K-means clustering algorithm can be used to group courses.Each course is represented as a feature vector, such as (course type, credits, hours, semester), and then the Euclidean distance is used to measure the similarity between courses. Through iterative optimization, the courses are divided into k groups, and the course characteristics in each group are highly similar. By observing the center point characteristics of each group, the distribution pattern of different types of courses can be found, such as public basic courses are mostly concentrated in lower grades, and professional courses are mostly concentrated in higher grades. Use the Apriori association rule mining algorithm to analyze the correlation between courses, such as {Advanced Mathematics, Linear Algebra}->{Probability Theory and Mathematical Statistics} means that students who have studied advanced mathematics and linear algebra are likely to study probability theory and mathematical statistics, thereby discovering common course combination patterns and providing references for course recommendations and academic planning.

[0020] Step S102, based on the school's curriculum data, determine the course availability information for each time period, and combine the school activity data to determine the course availability and the demand for the use of the activity venue during the activity period.

[0021] By analyzing the course schedule data, the number and distribution information of courses in each time period are counted, the course occupancy rate of each time period is calculated, and the course idleness of different time periods is determined according to the course occupancy rate, so as to obtain a time occupancy heat map of the course idleness information of the whole school; the important activity arrangements in the school calendar are obtained, the time, location and type of the activity arrangements are extracted, and the use demand of the activity venue is judged; the time information of the school activity data is superimposed and analyzed with the time occupancy heat map, the tension index of the venue resources in each time period during the activity is calculated, and the peak time period of the activity and the hot time period of each venue are identified; the occupancy information of each classroom in different time periods is obtained by statistical analysis of the classroom information in the course schedule data of the whole school, and the idle classrooms in each time period are analyzed to determine whether the idle classrooms meet the use demand of the activity venue; the reservation and use information of the sports venue is obtained, and the venue occupancy rate and idleness of each time period are obtained in combination with the course schedule data and the school activity data, and the venue scheduling suggestions are generated to ensure that the use demand of the activity venue is met by utilizing the idle time period under the premise of ensuring physical education teaching;

[0022] For example, by analyzing the course schedule data of the whole school, counting the number and distribution of courses in each time period, dividing a day into several time periods (such as morning, afternoon, and evening), marking the time periods with different colors according to the number of courses, and using Python's matplotlib library or JavaScript's D3.js library to generate a time occupancy heat map that reflects the free situation of courses in the whole school. Obtain the important event schedules in the school calendar, extract the time, location, type and other information of the event, and judge the potential use demand for the event venue according to the type of activity. For example, sports competitions, sports meetings and other activities require a large number of sports stadiums, while tea parties and other activities only need to use idle classrooms. The time information of the activity is superimposed and analyzed with the course idleness heat map. The activity arrangement information is divided into time periods and aligned with the time periods of the course idleness heat map. For each time period, the number of activities and course idleness are counted, and the tension index of the venue resources is calculated. According to the tension index, the time periods are sorted and marked to obtain the superimposed analysis results, and the tension of venue resources in each time period during the activity is judged. The peak time period of the activity and the time period with the most intense competition for venue resources are identified, which provides a basis for the rational allocation of venue resources. By statistically analyzing the classroom information in the course arrangement data, the occupancy of each classroom in different time periods is obtained, and the number of available classrooms in each time period is summarized to determine whether the idle classrooms can meet the demand for activity venues, and the course classrooms and activity venues are planned in a coordinated manner. The reservation and use of sports venues are obtained from the venue resource management system. Combined with the course arrangement and activity arrangement, the use of sports venues is analyzed to obtain the venue occupancy rate and idleness of each time period, and the feasibility of using idle time periods to meet the demand for activities under the premise of ensuring physical education teaching is determined. Taking into account the availability of courses, classroom occupancy and sports venue occupancy, the DataCube model is used to combine dimensions such as courses, classrooms, and time periods to form a multidimensional data set. The multidimensional data set is sliced, diced, and pivoted to generate summary data of different dimensions. The association rule mining algorithm (such as the Apriori algorithm) is applied to discover the association patterns between different dimensions, and the intensity of various venue resources during the event is obtained. The venue resources in different time periods are optimized, such as using lunch breaks and after-school time to arrange sports activities and flexibly switching between teaching buildings and stadiums. For large-scale events and competitions, the demand for venue resources is predicted and evaluated in advance, and space is freed up to the maximum extent by optimizing course arrangements and adjusting teaching locations. At the same time, emergency plans are formulated to ensure the smooth development of activities. In the course arrangement data analysis, a day can be divided into 8 time periods, each of which is 90 minutes. The number of courses in each time period is counted, and the heat map is drawn using Python's matplotlib library.In specific implementation, the number of courses can be divided into 5 levels, represented by dark blue, light blue, green, orange and red, and the corresponding range of the number of courses is 0-10, 11-20, 21-30, 31-40 and greater than 40. By observing the distribution of different colors in the heat map, the free situation of the whole school's course arrangement can be intuitively grasped. In the superposition analysis of activity arrangement and course free situation, the activity arrangement information and the course free situation heat map can be divided according to the same time period. For each time period, the ratio of the number of activities to the course free degree is calculated to obtain the site resource tension index. For example, there are 2 activities arranged in a certain time period, and the course free degree is 80%, then the tension index is 2 / 0.8=2.5. Sort the tension index of each time period, take the top 20% of the time period as the activity peak period, and the corresponding tension index threshold is 3.0. In the multi-dimensional data association analysis, the dimensions of course, classroom, time period, etc. can be combined to form a data cube, and each cell represents the course arrangement of a classroom in a certain time period. By slicing the data cube, we can get the course schedule of different classrooms; by dicing, we can get the course schedule of all classrooms in a specific time period; by rotating, we can adjust the order of dimensions and generate summary data from different angles. When applying the Apriori algorithm for association rule mining, we can treat the usage of each classroom in each time period as a transaction, and find frequent item sets and association rules by setting the minimum support and minimum confidence. For example, the support of "classroom A is free in both the first and second classes in the morning" is 85%, and the confidence of "classroom A is free in the first class in the morning" and "classroom A is free in the second class in the morning" is 95%, indicating that the idle status of these two time periods has a strong correlation, and activities held continuously in these two time periods can be arranged first.

[0023] Step S103, setting up a prediction model based on time series prediction or regression analysis, inputting the school's course schedule, school activities and historical venue usage data, outputting the predicted value of venue usage rate in each time period by training and optimizing the prediction model parameters, and identifying the expected peak period.

[0024] Collect and organize the school's course schedule, school activities, and historical venue usage data, clean, standardize, and format the data, extract key features related to venue usage through feature engineering, including course type, activity type, and time period, and build a structured training data set. Set up a prediction model based on time series prediction or regression analysis, including long short-term memory neural network (LSTM) or temporal convolutional network (TCN), and determine the input and output format, network structure, and hyperparameters of the prediction model based on the characteristics of the training data set and the prediction target. Tune the hyperparameters through grid search or Bayesian optimization, including tuning the learning rate, batch size, number of hidden layers, and number of neurons. Optimize the parameters of the prediction model through cross-validation.

[0025] For example, the LSTM model can adopt a multi-layer stacked structure, each layer contains several LSTM units, and the gating mechanism is used to achieve the modeling of long-term and short-term dependencies; the TCN model can use technologies such as causal convolution and dilated convolution to expand the receptive field and maintain the causality of the time dimension, while accelerating the convergence speed through residual connections. The training data set is divided into a training set, a validation set, and a test set in chronological order, and the model is trained using the training set. The predicted value is calculated by forward propagation, the model weight is optimized by back propagation, and the model performance is evaluated on the validation set. The model structure and hyperparameters are adjusted according to the evaluation results, and the training process is continuously iterated until the model performance reaches the expected goal or converges. The trained prediction model is tested and evaluated, and the predicted value is calculated using the test set data. The prediction accuracy is measured by indicators such as mean square error (MSE) and mean absolute error (MAE). For error indicators such as MSE and MAE, the smaller the value, the smaller the deviation between the predicted value and the true value, and the higher the prediction accuracy. The mean and variance of the error indicators can be calculated, the distribution characteristics of the prediction error can be analyzed, and the time periods with better and worse prediction effects can be identified. In addition to the error index, the coefficient of determination (R-squared), mean absolute percentage error (MAPE) and other indicators can also be used to evaluate the goodness of fit and relative error level of the prediction model. According to the evaluation results, the model structure, feature selection and training strategy are further optimized to improve the overall prediction level. Apply the trained prediction model, input the course schedule and school activity data for the next period of time, and obtain the predicted value of venue utilization rate in each time period through forward propagation calculation. Post-process the predicted value, such as moving average smoothing, select a suitable window size, such as 3 or 5 time steps, calculate the arithmetic mean of the predicted value in the window, and obtain the smoothed prediction curve; use wavelet transform to denoise the predicted value, by selecting the appropriate wavelet basis function and decomposition layer, multi-scale decomposition of the predicted value, remove high-frequency noise components, and reconstruct the denoised prediction curve; combine the predicted value and historical data, and use Kalman filtering and other methods to dynamically adjust and correct, so as to obtain more stable and reliable prediction results. According to the peak and change trend of the prediction curve, the venue usage in each time period in the future can be judged, the expected peak and trough periods can be identified, and the time pattern and cause of the peak period can be analyzed to provide a decision-making basis for venue management and resource scheduling, and to issue an early warning for the time period when the venue usage rate exceeds the threshold. The training data set should be updated regularly, and the latest course schedule, school activities and venue usage data should be included in the model training. Through incremental learning or regular retraining, the prediction model can be adapted to the changing trend of the data, improve the stability and adaptability of the prediction model, and provide support for long-term venue usage rate prediction. In the data preprocessing stage, Python's Pandas library can be used to clean and convert the data, such as converting date and time strings to timestamp format, filling missing values ​​with the average of the previous and next time periods, and replacing outliers with the median.Through feature importance analysis methods, such as feature importance evaluation based on tree models or recursive feature elimination, the top-10 features with the highest correlation with venue utilization rate are selected as model input. In the design of the LSTM model, a 3-layer stacked structure can be used, each layer contains 64 LSTM units, using the tanh activation function and a dropout ratio of 0.2, the optimizer selects Adam, and the learning rate is set to 0.001. In the design of the TCN model, 2 convolutional layers can be used, each layer contains 128 convolution kernels, the convolution kernel size is 3, the void rates are 1 and 2 respectively, the ReLU activation function and a dropout ratio of 0.1 are used, the optimizer selects SGD, and the learning rate is set to 0.01. During the model training process, MSE and MAE are evaluated on the validation set after each epoch. When the validation set error does not decrease for 5 consecutive epochs, the early stopping mechanism is triggered to save the model parameters with the best performance. In the model testing phase, the test set data is generated using a sliding window method, with a window size of 7 days and a step size of 1 day. The last day of each window is predicted. The predicted results were compared with the true values, and the calculated MSE was 0.05, MAE was 0.08, MAPE was 6%, and R-squared was 0.92, indicating that the model had good prediction performance. In the post-processing stage, the predicted values ​​were smoothed by moving average, and the window size was selected as 3 time steps, that is, the average of the current time step and the three predicted values ​​before and after was taken as the smoothing result. The predicted values ​​were decomposed into three layers using Haar wavelet, and the high-frequency detail coefficients were removed to reconstruct the denoised prediction curve. Finally, combined with the historical site utilization data, Kalman filtering was used for dynamic tracking and correction, and the noise covariance matrix was estimated to be a diagonal matrix. The state transfer matrix and observation matrix were estimated based on the autoregressive coefficients of the historical data to obtain the final prediction results. The prediction results show that the average site utilization rate in the next week is 65%, and the peak period occurs in the afternoon of Wednesday and Friday, with the highest utilization rate reaching 85%. Resources and personnel need to be deployed in advance to cope with peak demand.

[0026] Step S104, obtaining the user's personal course schedule data and extracting the user's free time data. At the same time, analyzing the user's exercise characteristics from the user's historical exercise data, including exercise type, exercise intensity and exercise habits, and establishing a user exercise characteristic table.

[0027] By connecting with the school's academic affairs system, the user's personal course schedule data is obtained, and the course schedule is extracted, including the course name, class time, and course duration. According to the course schedule, the user's free time data is determined to form the user's free time schedule; from the user's historical exercise records, the types of sports that the user has participated in, including basketball, football, and swimming, are obtained, and the number of times and duration of participation in each type of sports are counted to obtain the user's preference for different types of sports, and the frequency and duration distribution of the user's participation in sports in different time periods are analyzed to determine the user's exercise time habits; according to the duration and frequency of the user's participation in different types of sports, the user's exercise intensity in each exercise is estimated, and the average speed is calculated through running distance and time, and the average speed is calculated through ball sports. The duration and intensity classification of the user are comprehensively evaluated to evaluate the user's consumption and exercise intensity level in different sports to obtain the user's exercise intensity characteristics; according to the user's exercise type preference, exercise time habit, and exercise intensity characteristics, the user's exercise characteristic table is constructed; the user's course free timetable and exercise time habit are analyzed to determine the time period when the user is accustomed to exercise in his free time, and in combination with the user's exercise type preference, the user's sports items of interest and preferred exercise time period are preliminarily screened out; according to the user's exercise intensity characteristics and physical health information, the user's sports items of interest are refined and adjusted, and in combination with the learned venue information, the user's exercise time period and exercise type are matched, the recommended sports venue corresponding to the user is determined, and a sports recommendation plan is generated.

[0028] Exemplarily, the user is generally a student, but can also be a faculty member. By connecting with the school's academic affairs system, the user's personal course schedule data is obtained, and information such as course name, class time, and course duration are extracted. According to the course schedule, the user's free time in each time period is judged, and the user's free time data is obtained to form a user free time table. From the user's historical sports records, the types of sports that the user has participated in, such as basketball, football, swimming, etc., are obtained. Assuming that the original data contains fields such as user ID, sports type, sports start time, and sports end time, tools such as SQL or Python are used to clean and convert the data, group them according to user ID and sports type, and count the number of times each user participates in each type of sports and the total duration. The statistical results are saved as a user sports preference table, which contains fields such as user ID, sports type, number of times, and total duration, and the user's preference for different types of sports is obtained. At the same time, the frequency and duration distribution of users participating in sports in different time periods are analyzed to judge the user's sports time habits. According to the duration and frequency of users' participation in different types of sports, the exercise intensity of users in each exercise is estimated. For aerobic exercises such as running and swimming, the calories consumed are estimated based on the duration and intensity of exercise. For example, 30 minutes of moderate-intensity running consumes about 300 calories. For anaerobic exercises such as strength training, the consumption is estimated using the EPOC (excess post-exercise oxygen consumption) formula based on factors such as the number of exercise groups, weight, and number of times. The consumption of aerobic and anaerobic exercises is combined to obtain an estimated value of the total consumption of users in various types of sports, evaluate the exercise intensity level of users in different sports, and obtain the user's exercise intensity characteristics. Summarize the user's sports type preference, exercise time habits, exercise intensity characteristics and other information, build a user sports feature table, and design the corresponding database table structure. The table name is user_sports_profile, which contains the fields user_id(varchar), sport_type(varchar), preference_score(float), usual_time_slot(varchar), frequency_per_week(int), average_duration(int), average_intensity(float). The primary key is user_id and sport_type. Create indexes on the user_id, sport_type, and usual_time_slot fields as the basic data for subsequent sports recommendations and matching. Analyze the user's free timetable and exercise time habits, find out the overlap between the two, determine the time period when the user is accustomed to exercising in his free time, and combine the user's sports type preference to preliminarily screen out the sports and time periods that the user may be interested in.According to the user's exercise intensity characteristics and physical condition, such as BMI index, physical fitness test results, etc., the initially screened sports items are further refined and adjusted. According to the user's BMI range (such as thin, normal, fat), different exercise types and intensity thresholds are set to score the candidate sports items. Factors considered include the user's preference for the exercise type, the matching degree of exercise intensity with the user's physical fitness, the overlap between exercise time and user's free time, etc., and the candidate items are sorted based on the scores. The top N with the highest scores are selected as the recommendation results. According to the user's feedback and actual participation, the recommendation model is updated and optimized in real time. Combined with the school's sports venue resources, such as the opening hours and free status of gymnasiums, swimming pools, playgrounds, etc., the user's exercise time period and exercise type are matched, and suitable sports venues and facilities are recommended. Finally, a personalized exercise recommendation plan is generated, including recommended exercise type, exercise time, sports venue, exercise intensity, etc., for user reference and selection. When obtaining user course schedule data, you can call the API interface of the academic affairs system, pass in the user's student number or authentication information, obtain the returned course schedule data in JSON format, and then use Python's json library to parse the data, extract fields such as course name, class time, and course duration, and build the user's free time table. For example, if the user has classes from 8:00-10:00 and 14:00-16:00 from Monday to Friday, mark the other time periods as free time. For the processing of historical sports data, you can use big data processing tools such as Hive or Spark, write SQL query statements, group by user ID and sports type, use COUNT(*) and SUM(duration) functions to calculate the number of times and total duration of each user's participation in each sport, and then use Python's Pandas library to read the query results and generate a user sports preference table. When calculating sports consumption, you can define a calculation function, such as defcalc_calories(sport_type,duration,intensity), and select different calculation formulas according to the type, duration, and intensity of sports. For example, for moderate-intensity running, you can use the formula calories = 0.17*weight*duration, where weight is the user's weight and duration is the running time in minutes. For strength training, you can use the formula calories = sets*reps*weight*0.033, where sets is the number of sets, reps is the number of times per set, and weight is the weight in kilograms.When constructing a user sports feature table, you can use the MySQL database to create a table named user_sports_profile, which contains fields such as user_id, sport_type, preference_score, usual_time_slot, frequency_per_week, average_duration, and average_intensity. The preference_score can be calculated based on the number of times and total duration of the user's participation in the sport type, usual_time_slot can be determined based on the frequency of the user's participation in the sport in different time periods, frequency_per_week and average_duration can be directly obtained from the statistical results, and average_intensity can be estimated based on the user's average consumption of the sport type. When generating personalized sports recommendations, you can first divide the user into three levels based on the BMI index: thin (BMI<18.5), normal (18.5≤BMI<24), and fat (BMI≥24). Then, for each level, set different sports types and intensity thresholds, such as recommending low-intensity aerobic exercise for thin users, medium-intensity comprehensive exercise for normal users, and high-intensity strength training and aerobic exercise for fat users. When screening candidate sports events, the comprehensive score of each event can be calculated, such as preference_score*0.4+intensity_match_score*0.3+time_match_score*0.3, where intensity_match_score is calculated based on the user's exercise intensity characteristics and the event's exercise intensity threshold, and time_match_score is calculated based on the user's free time and the event's opening time. Finally, the top 3 with the highest comprehensive scores are selected as recommendation results to be displayed to the user.

[0029] Step S105, based on the user's exercise feature table and the venue peak period, collaborative filtering or content-based recommendation model is used to generate personalized exercise time recommendations for each user, including obtaining the user's optimal exercise time based on the user's free time, exercise features and venue peak period.

[0030] Extract the sports preference, sports intensity, and sports time of each user's sports characteristics from the user sports feature table, obtain the congestion level of different sports venues in each time period from the peak data table, and construct a user-time period-venue triple data structure. Calculate the similarity of sports preferences between different users, determine the k neighbor users with the highest similarity to the target user's sports preferences, and select the k neighbor users' sports time periods as candidate sports time periods. Analyze the project features of each sports project, including sports type, sports intensity, and required venues, construct a knowledge graph of sports projects, form a semantic feature vector, and calculate the similarity between the user feature vector and the sports project feature vector. Combine the user's free time, sports preference, and venue congestion level, score and sort the candidate sports time periods, and select the candidate sports time period with the highest score as the best sports time.

[0031] Exemplarily, the features of each user's sports preference, sports intensity, sports time period, etc. are extracted from the user sports feature table, and the congestion degree of different sports venues in each time period is obtained from the peak data table. These data are loaded into the input data set of the recommendation algorithm to construct a user-time period-venue triple data structure. For the collaborative filtering algorithm, the similarity of sports preferences between different users can be calculated using a measurement method such as the Pearson correlation coefficient or cosine similarity. The specific calculation formula is as follows: Pearson correlation coefficient sim(u,v)=Σ(Rui-Ru)(Rvi-Rv) / sqrt(Σ(Rui-Ru)^2Σ(Rvi-Rv)^2), cosine similarity sim(u,v)=Σ(Rui*Rvi) / sqrt(Σ(Rui^2)Σ(Rvi^2)), where Rui and Rvi represent the scores of users u and v for sports item i, respectively, and Ru and Rv represent the average scores of users u and v. According to the user's historical exercise data, find the k neighbor users who are closest to the target user's preferences, and select their exercise time periods as candidate recommendations. For content-based recommendation algorithms, analyze the characteristics of each sport, such as exercise type, exercise intensity, required venue, etc., build a knowledge graph of sports, including entities and relationships such as exercise type, equipment, venue, body part, and exercise effect. For each sport, extract its related entities and relationships from the knowledge graph to form a semantic feature vector, and calculate the similarity between the user feature vector and the sport feature vector as the basis for recommendation scores. When generating exercise time recommendations, comprehensively consider factors such as the user's free time, exercise preferences, and venue congestion, and score and sort the candidate time periods. Combination strategies such as weighted average can be used to determine the weights of each factor in the following ways: According to historical data, count the correlation between different factors and user acceptance of recommendations, such as click-through rate, satisfaction, etc.; use machine learning algorithms such as linear regression and gradient boosting tree to train weight models and automatically learn the optimal weight combination of each factor; for new users or sparse data, heuristic weight settings can be used, such as initialization based on domain knowledge and expert experience. The time period with the highest comprehensive score is obtained as the recommendation result. In order to improve the diversity and novelty of recommendations, a certain random exploration mechanism can be introduced to add a certain proportion of non-popular or previously untried sports time periods to the recommendation results, evaluate the effect of exploratory recommendations through AB testing and other methods, and dynamically adjust the exploration ratio. Based on user feedback on the recommendation results, such as acceptance, rejection, and rating, collect users' implicit or explicit preference data, update the user's sports feature table and the similarity matrix of the collaborative filtering algorithm, and regularly retrain the recommendation model to achieve continuous optimization and personalized adjustment of the recommendation results.Evaluate and monitor the effect of the recommendation algorithm. Use offline and online indicators to comprehensively judge the recommendation quality and user experience of the algorithm. The specific calculation method is as follows: Precision = TP / (TP+FP), Recall = TP / (TP+FN), Coverage = Number of recommended items / Total number of items, Click-through rate CTR = Number of clicks / Number of impressions, Acceptance rate Acceptance rate = Number of acceptances / Number of recommendations, where TP represents true positives, FP represents false positives, and FN represents false negatives. Continuously iterate and optimize algorithms and parameters to ensure continuous improvement of recommendation results. When constructing the user-time period-venue triple data structure, a representation method similar to the knowledge graph can be used, with users, time periods, and venues as entities, and the relationships between them such as "like" "free time is" "crowded degree is" as relationship edges. By establishing nodes and edges in graph databases such as Neo4j, the data is converted into a graph for storage. For the calculation of user similarity, taking the ratings of two users on 10 sports as an example, the process of using the Pearson correlation coefficient is as follows: first calculate the average rating of each user, such as the average rating of user A is 4.2 and that of user B is 3.8; then for each jointly rated item, calculate (Rui-Ru)(Rvi-Rv), such as user A rated 4 on item 1 and user B rated 3, then (4-4.2)(3-3.8) = 0.16, and so on, sum and divide by the square root of the product of the denominator to finally get the similarity value. When constructing a sports knowledge graph, taking football as an example, we can extract relationship triplets such as "football-ball sports", "football-outdoor sports", "football-grass", and "football-sports events". Based on knowledge graph representation learning algorithms such as TransE, we can vectorize the relationship triplets through methods such as random walks and negative sampling, and each entity and relationship corresponds to a low-dimensional dense vector. When generating recommendation results, the weight of each factor can be set. For example, the weights of user preference, free time matching, and venue congestion are set to 0.5, 0.3, and 0.2 respectively. For a user who wants to play basketball at 3 pm on Wednesday, assuming that the user preference for Basketball is 4.5, the free time matching at 15:00 on Wednesday is 0.8, and the basketball court congestion is 0.4, the final score is 4.50.5+0.80.3+0.40.2=2.57. When evaluating the recommendation effect, the accuracy and recall rate can be calculated by constructing a confusion matrix. For example, the recommendation results can be divided according to acceptance / rejection and actual participation / non-participation, so as to obtain the number of TP, FP, TN, and FN. Substituting them into the formula can obtain the accuracy and recall values, and then analyze the advantages and disadvantages of the recommendation algorithm.

[0032] If the user's free time coincides with the venue's peak time, the recommended optimal exercise time is adjusted according to the user's exercise intensity and exercise duration. If the user's free time is greater than the threshold, the recommended optimal exercise time is not adjacent to the venue's peak time.

[0033] Obtain the user's free time data and venue peak data, convert the user's free time and venue peak into time segments, and calculate the overlap of time periods. If the user's free time coincides with the venue peak, extract the user's exercise intensity and exercise duration, and calculate the exercise intensity interval and exercise duration interval based on the exercise intensity and exercise duration. During the user's free time, select the time period adjacent to the venue peak and whose stress index is lower than the threshold as the candidate recommended time period. Screen the candidate recommended time periods based on the user's exercise intensity interval and exercise duration interval, calculate the similarity, and select the candidate recommended time period with the highest similarity as the best exercise time. If the user's free time is greater than the threshold, select the best exercise time from the non-peak time period, and the difference between the non-peak time period and the venue peak time is greater than the preset threshold.

[0034] Exemplarily, the user's free time data and venue peak period data are obtained, and the use of interval trees or segment trees and other data structures is considered to convert the user's free time and venue peak period into segments. Through the query and update operations of the tree structure, the overlap of time periods is efficiently calculated. According to historical data and business experience, a reasonable overlap threshold is set, such as 30% or 50%, to balance the coverage and accuracy of recommendations. If the overlap exceeds the threshold, it is determined that the user's free time and venue peak period have significant overlap. If the user's free time coincides with the venue peak period, the user's exercise intensity and exercise duration preference data are extracted, and according to the distribution of exercise intensity and duration, the user's habitual exercise intensity interval and exercise duration interval are calculated as the screening conditions for the recommended time period. During the user's free time, select a time period that is close to the peak period but less stressful as a candidate recommended time period. The stress index can be measured by the average crowdedness of the venue, the flow of people and other indicators. The specific calculation formula is: stress index = venue flow of people / maximum capacity of the venue. The flow of people data can be collected in real time by counters or cameras at the entrance of the venue. The maximum capacity can be estimated based on the area of ​​the venue, the number of equipment, etc. When calculating the stress index, a certain time window, such as 10 minutes or 30 minutes, can be considered to smooth instantaneous fluctuations, reflect the overall stress level within the time period, and select a time period with a stress index below a certain threshold. According to the user's exercise intensity interval and exercise duration interval, the candidate recommended time periods are screened. Using common similarity measurement methods such as cosine similarity or Euclidean distance, the user preference and time period features are represented as vectors, and the similarity between them is calculated. The user preference vector = [exercise intensity, exercise duration], the time period feature vector = [average exercise intensity, average exercise duration], the cosine similarity = vector dot product / (vector modulus product), and the Euclidean distance = the L2 norm of the vector difference. By setting a similarity threshold or sorting, the time period with the highest similarity is selected as the recommendation result. If the user has sufficient free time, that is, the length of the free time is significantly longer than the user's average exercise time, priority will be given to recommending time periods that avoid peak hours, and selecting the time period with the highest similarity in exercise intensity and duration from the non-peak time periods as the recommendation result, with reference to the similarity calculation mentioned above. The difference between the non-peak time period and the venue peak time is greater than 10 minutes.In the process of selecting the recommended time period, the continuity and stability of the time period are comprehensively considered. The continuity of the time period can be measured by the interval between the time periods. If the interval exceeds a certain threshold (such as 30 minutes), the time period is considered discontinuous; the stability of the time period can be measured by the frequency and variance of the time period in different cycles (such as weekly and monthly). The higher the frequency and the smaller the variance, the more stable the time period is considered; for the judgment of periodicity, time series analysis methods such as Fourier transform or autocorrelation analysis can be used to extract the periodic components of the time period series. For example, for the recommendation of fixed time periods every week, continuous time periods should be selected as much as possible to avoid frequent switching and interruptions. At the same time, the predictability of the time period should be considered, and time periods with periodicity should be given priority. For different user groups, different recommendation strategies and parameters can be set according to the user's age, gender, occupation and other attributes. For example, for the student group, the time periods between classes and after class can be given priority; for the student group, the time periods after class and weekends can be given priority to improve the personalization and accuracy of the recommendation. When calculating the overlap between the user's idle time and the venue's peak period, the time period can be represented as an integer interval, such as 8:00-9:00 represented as [800,900), and then stored and queried using an interval tree. For each user's idle time interval, search on the interval tree to find all the venue's peak period intervals that intersect with it, add up their intersection lengths, and divide by the total length of the user's idle time to get the overlap. For example, the user's idle time is [800,1000) and [1400,1800), and the venue's peak period is [830,930) and [1600,1730), then the overlap is (30+30) / (200+400)=0.1. If the set overlap threshold is 0.3, then the user's idle time does not overlap significantly with the venue's peak period. When generating the recommended time period, the user preference vector and time period feature vector can be constructed according to the user's exercise intensity and duration preferences. For example, if the user preference is [0.6,90], it means that the user likes an exercise intensity of 0.6 and an exercise duration of 90 minutes. The feature of a time period is [0.7,60], which means that the average exercise intensity of the time period is 0.7 and the average exercise duration is 60 minutes. The cosine similarity is used to calculate the similarity of the two vectors. The formula is cosθ=(0.60.7+9060) / (sqrt(0.6^2+90^2)sqrt(0.7^2+60^2))=0.832, indicating that the match between the time period and the user preference is 0.832. The matching degree is calculated for all candidate time periods, and the top N with the highest scores are selected as the recommendation results. When judging the continuity of the time period, the interval distribution between the recommended time periods can be counted. If more than 80% of the intervals are less than 30 minutes, it is considered that the overall continuity of the recommendation results is good.When judging the stability of a time period, the number of times each time period appears in different cycles can be counted. If a time period appears every week in the past four weeks and the time deviation of each occurrence is less than 15 minutes, then the time period is considered to have strong stability and periodicity. When setting recommendation strategies for different user groups, users can be divided into different groups through clustering algorithms, such as using the K-Means algorithm to cluster based on user characteristics such as age, gender, occupation, and sports preferences to obtain K user groups. For each group, the free time distribution, exercise intensity, and duration preferences of its members are counted separately to find the differences in time preferences among different groups, so as to set targeted recommendation strategies and improve the overall recommendation effect.

[0035] Step S106, regularly collect user feedback data, including user satisfaction with the optimal exercise time, the difference between the actual exercise time and the optimal exercise time, and update the user exercise feature table based on the feedback data.

[0036] Acquire the feedback data, perform data cleaning, deduplication and formatting on the feedback data, extract key fields, and construct a structured feedback data set; perform statistical analysis on the feedback data set, calculate the average satisfaction scores of different time periods, draw a satisfaction distribution histogram, and determine the time period when the average satisfaction score is lower than a threshold; and calculate the average deviation and deviation distribution between the actual exercise time and the optimal exercise time, and determine the time period when the average deviation is higher than the threshold; for the time period when the average satisfaction score is lower than the threshold and the average deviation is higher than the threshold, identify key features through feature selection and correlation analysis, adjust the weights and thresholds of the key features, and modify the parameters of the recommendation model; dynamically adjust the parameters of the recommendation model according to the user's satisfaction score for the optimal exercise time; regularly update the actual exercise time and satisfaction score in the user's feedback data to the user's exercise feature table to replace or supplement the original exercise feature value.

[0037] Exemplarily, a user feedback data collection system is designed and developed. By embedding a satisfaction rating control and a feedback submission form on the recommendation result page, the user is guided to rate the satisfaction of the recommended time after each recommendation and completion of the exercise. The satisfaction rating can adopt a 5-level scale, namely very dissatisfied, dissatisfied, general, satisfied, and very satisfied, corresponding to 1-5 points respectively. The feedback submission form can include fields such as the recommended time period, the actual exercise start time, and the actual exercise end time, as well as an open text box for the user to fill in other feedback opinions. After the user completes the rating and feedback submission, a confirmation dialog box can pop up to prompt the user that the feedback has been successfully recorded and thank the user for his valuable opinions. The collected user feedback data is cleaned, deduplicated, and formatted, and key fields such as user ID, recommended time period, satisfaction rating, actual exercise start time, actual exercise end time, etc. are extracted to construct a structured feedback data set, which is stored in the feedback database and associated with the user exercise feature table. The feedback data set is statistically analyzed to calculate the average satisfaction scores of different time periods and different user groups, and a satisfaction distribution histogram is drawn. For satisfaction rating, a lower threshold can be set, such as 3 points. For time periods or user groups with an average satisfaction score below 3 points, they are marked as having low satisfaction. The average deviation and deviation distribution of the actual exercise time and the recommended time can be calculated. For the deviation of the actual exercise time and the recommended time, an upper threshold can be set, such as 30 minutes. For time periods or user groups with an average deviation greater than 30 minutes, they are marked as having large deviation. The setting of the threshold can be adjusted according to historical data and business needs, or a dynamic threshold can be used to adaptively adjust according to the distribution of feedback data. Identify time periods and user groups with low satisfaction and large deviations as optimization priorities. For time periods and user groups with low satisfaction and large deviations, analyze their feature distribution and historical recommendation data, use methods such as chi-square test and information gain to calculate the correlation between each feature and satisfaction and deviation, and select the top-N features with the highest correlation as key features; you can also use methods such as recursive feature elimination (RFE) and L1 regularization to recursively train the model and evaluate the importance of features to gradually filter out the optimal feature subset. After feature selection, use indicators such as Pearson correlation coefficient and mutual information to calculate the correlation between key features and satisfaction and deviation, quantitatively analyze the influence of features, adjust the weights and thresholds of these features, and modify the corresponding parameters of the recommendation algorithm. Add a feedback learning mechanism of user feedback data to the recommendation algorithm, and use online learning algorithms such as stochastic gradient descent (SGD) and AdaGrad to update model parameters in real time according to each user's feedback data. Specifically, use the satisfaction score of user feedback as the label of the sample, and the characteristics of the recommended time period as the characteristics of the sample to build a satisfaction prediction model.After receiving user feedback each time, it is used as a new sample, and the model parameters are updated using the online learning algorithm so that the model can quickly adapt to changes in user preferences. When dynamically adjusting the recommendation model, a decay factor is set to control the weight of historical feedback data, so that the model pays more attention to recent feedback data, and the recommendation results can continuously adapt to changes in user preferences. Regularly update the actual exercise time and satisfaction score in the user feedback data to the user exercise feature table, replace or supplement the original feature values ​​such as exercise duration preference and exercise intensity preference, so that the user feature table can timely reflect the user's actual exercise situation and preference changes, and provide more accurate input for subsequent recommendations. Establish an evaluation and monitoring mechanism for the recommendation algorithm, use the feedback data set as a test set, regularly calculate the accuracy, recall rate, satisfaction and other indicators of the recommendation results, track the trend of indicators, promptly discover the decline or abnormality of the recommendation effect, conduct cause analysis and algorithm optimization, and ensure the continuous improvement of the recommendation algorithm and the continuous improvement of user experience. In the user feedback data collection system, the satisfaction rating control can be represented by 5 star icons. Users can complete the rating by clicking the corresponding star icon, and the backend converts the number of stars into a numerical value of 1-5 points for storage. The feedback submission form can be implemented using a web form, using the HTML5 datetime-local type input tag to obtain the actual exercise start and end time selected by the user, and using the textarea tag to obtain other user feedback. In the data cleaning and formatting stage, regular expressions can be used to verify and standardize the time format, such as converting "2023-04-2018:30" to a unified "2023-04-20T18:30:00" format. In the statistical analysis stage, Python's numpy library can be used to calculate the average satisfaction score and deviation value, and the matplotlib library can be used to draw a histogram. For example, if the satisfaction score list is [3,4,5,2,4], the average satisfaction is 3.6, and the proportion below the threshold of 3 is 20%. In the feature selection stage, the SelectKBest class of the sklearn library can be used to implement the chi-square test and select the top-5 features with the highest chi-square statistics. The chi-square test found that "exercise intensity preference" and "weather conditions" have the highest chi-square statistics with satisfaction, so they are included in the recommendation algorithm as key features. In the feedback learning mechanism, the SGDClassifier class of the sklearn library can be used to implement the stochastic gradient descent algorithm, taking each user's feedback data as a sample, the satisfaction score as a label, and the recommended time period characteristics as a feature vector to update the model parameters in real time.When receiving feedback data from a user, extract the recommended time period features [0.8, 60, 1] (exercise intensity, duration, whether it is a weekend), and the satisfaction score is 4, then input it as a positive sample [0.8, 60, 1, 1] into SGDClassifier for training and update model parameters. In the evaluation and monitoring stage, the classification_report function of the sklearn library can be used to calculate the accuracy, recall, F1 value and other indicators of the recommendation results. Evaluate the recommendation model on the test set, and get an accuracy of 85%, a recall of 80%, and an F1 value of 82%, which means that the current model has good performance and can continue to be used. If the accuracy is found to be lower than 70% or the recall is lower than 60%, algorithm optimization and parameter adjustment are required.

[0038] Regularly collect users’ satisfaction with the optimal exercise time and the deviation data between the actual exercise time and the optimal exercise time through online questionnaires and in-app feedback functions.

[0039] Regularly collect user feedback texts through online questionnaires and in-application feedback functions, identify satisfaction with the optimal exercise time and deviation data between the actual exercise time and the optimal exercise time in the feedback text; perform natural language processing on the feedback text, including word segmentation, removal of stop words, and part-of-speech tagging; perform semantic analysis and sentiment analysis on the feedback text, and identify key topics and sentiment tendencies in the feedback text through word frequency statistics or TF-IDF weight calculation; perform descriptive statistical analysis on the quantitative data, calculate the mean, median, and standard deviation of the satisfaction score, and draw a satisfaction distribution histogram; calculate the absolute deviation and relative deviation of the deviation data between the actual exercise time and the optimal exercise time, and analyze the distribution characteristics of the deviation; based on the analysis results of the user satisfaction and deviation data, identify the time period when the satisfaction is lower than the threshold or the deviation is higher than the threshold, and combine the feedback text to identify the key factors affecting user satisfaction and deviation.

[0040] For example, an online questionnaire is preset and published, and the questionnaire content includes the user's satisfaction rating of the best exercise time (such as using a 5-point scale), as well as the filling of the actual exercise start time and end time. Use text analysis tools to process open feedback, and apply statistical analysis methods to parse quantitative data; immediately trigger a feedback pop-up window after the user completes a recommended exercise, and guide the user to fill in the questionnaire; after the user uses the recommendation function for one week, push a questionnaire filling reminder through an in-app message, and the frequency can be set to once a month, and set a certain reward mechanism, such as points or coupons, to increase user participation. Set a feedback entry in the application, such as adding a "Feedback" button on the recommendation result page, and a feedback pop-up window will pop up after clicking it. The size of the pop-up window can be adaptive according to the content, and it does not completely cover the main interface, so that users can close it at any time. The pop-up window content includes a satisfaction rating (such as using icons such as smiley faces, neutral faces, and crying faces) and an open feedback input box. The layout is concise and clear, avoiding too many words or operation steps. Set a "Submit" button at the bottom of the pop-up window, which will automatically close the pop-up window after clicking it, and give a thank you prompt to enhance user participation. Preprocess the collected feedback data. For quantitative data such as satisfaction ratings and actual exercise time, perform data cleaning, format conversion, and standardization to convert them into structured numerical data. For open-ended feedback text, use the nltk.tokenize module of the NLTK library to perform text segmentation, use the stop word list of the nltk.corpus module to filter stop words, and use the nltk.tag module to perform natural language processing steps such as part-of-speech tagging to extract keywords and sentiment tendencies. Use text analysis tools such as Python's NLTK library to perform semantic analysis and sentiment analysis on open-ended feedback text. Use the FreqDist function of the nltk.probability module to perform word frequency statistics, and use the TfidfVectorizer class of the sklearn library to perform TF-IDF weight calculation to identify key topics in user feedback. Use the SentimentIntensityAnalyzer class of the nltk.sentiment.vader module to perform sentiment analysis, obtain sentiment polarity and intensity, and perform visual display. Statistical analysis methods were used to perform descriptive statistical analysis on the quantitative data. The mean(), median(), and std() functions of the numpy library were used to calculate the mean, median, standard deviation and other indicators of the satisfaction score. The hist() function of the matplotlib library was used to draw a satisfaction distribution histogram. The boxplot() function was used to draw a satisfaction distribution boxplot to intuitively display the overall distribution of user satisfaction. For the deviation data between the actual exercise time and the optimal exercise time, the describe() function of the pandas library was used to calculate the descriptive statistical indicators of the deviation, such as the minimum, maximum, mean, quartile, etc., to analyze the distribution characteristics of the deviation.According to the analysis results of user satisfaction and deviation data, a lower threshold (such as 3 points) is set for the satisfaction score, and the user group or time period with an average satisfaction below the threshold is marked as having low satisfaction; an upper threshold (such as 30 minutes) is set for the deviation between the actual exercise time and the optimal exercise time, and the user group or time period with an average deviation above the threshold is marked as having a large deviation. The setting of the threshold can be adjusted according to historical data and business needs, or a dynamic threshold can be used to adaptively adjust according to changes in data distribution. Combined with open feedback content, summarize the key factors affecting user satisfaction and deviation, such as exercise intensity, duration, venue facilities, etc., to provide data support and improvement direction for subsequent optimization of recommendation algorithms. Generate feedback data analysis reports regularly, including statistical indicators of satisfaction and deviation data, key themes and emotional distribution of open feedback, analysis of influencing factors, etc., present the analysis results through data visualization for reference by algorithm engineers and product managers, continuously track the changing trend of feedback data, and evaluate the optimization effect of recommendation algorithms. When designing an online questionnaire, you can use online questionnaire tools such as GoogleForms or SurveyMonkey, set the satisfaction score as a required item, and use a 5-point scale, with 1 indicating very dissatisfied and 5 indicating very satisfied. The actual exercise time can be filled in with a date and time picker to avoid manual format errors by users. In the design of the in-app feedback pop-up window, the satisfaction score can use 5 emoticons, from angry to smiling, corresponding to 1-5 points, and the user can complete the score by clicking the corresponding icon. The feedback input box can limit the number of words, such as within 200 words, and the submit button can be set to a bright color, such as orange or blue, to guide users to click. In the data preprocessing stage, Python's pandas library can be used to read quantitative data, the dropna() function can be used to delete missing values, the astype() function can be used to convert the satisfaction score to an integer, and the to_datetime() function can be used to convert the actual exercise time to a date and time format. For text data, regular expressions can be used to filter out emoticons and special characters, and the jieba library can be used for Chinese word segmentation. In the text analysis stage, the FreqDist function of the NLTK library can be used to count the word frequency and generate a word cloud diagram. For example, the words "sports", "time", and "intensity" have the highest frequency, indicating that these are the focus of user feedback. Use the TextBlob library for sentiment analysis to obtain the sentiment polarity value of the feedback text, such as 0.2 for positive sentiment and -0.1 for negative sentiment. In the statistical analysis stage, the stats module of the scipy library can be used to test the normality of the satisfaction score. For example, the shapiro() function is used to calculate the p value. When p>0.05, the data is considered to follow the normal distribution and parametric statistical methods can be used; when p≤0.05, the data is considered to not follow the normal distribution and non-parametric statistical methods need to be used.Use the ttest_1samp() function of the stats module of the scipy library to perform a single-sample t-test to determine whether the mean of the satisfaction score is significantly higher or lower than the preset value, such as 3 points. In threshold judgment, you can set thresholds based on historical data and experience values, such as using the mean of the satisfaction score of 3.5 points as the lower threshold, and the median of the deviation between the actual exercise time and the optimal exercise time of 30 minutes as the upper threshold. Use the query() function of the pandas library to filter out data that meets the conditions for key analysis and improvement. When generating analysis reports, you can use JupyterNotebook to create interactive reports, use the Matplotlib library to draw histograms and line graphs of satisfaction scores, use the WordCloud library to generate word cloud diagrams of feedback keywords, and use the Seaborn library to draw bar graphs of sentiment polarity to intuitively display the analysis results.

[0041] Combined with the activity data recorded by the user's terminal and the behavior log in the application, it evaluates whether the user exercises according to the recommended content. The recommended content includes the best exercise time and corresponding sports events, and identifies the user's exercise pattern and frequency of participation in activities.

[0042] The embedded SDK of the application obtains the GPS positioning data of the user's mobile phone and collects the user's location information; the built-in accelerometer sensor of the terminal is used to record the user's motion acceleration data and obtain the user's motion trajectory and motion intensity information; a behavior log collection module is set in the application to obtain the interaction data between the user and the recommended content and form a structured behavior log database; the collected GPS positioning data is cleaned and preprocessed, and the GPS data is smoothed using the Kalman filter algorithm; the accelerometer data is normalized to extract the user's motion acceleration characteristics; the preprocessed GPS positioning data and the accelerometer data are fused with the behavior log data to construct a user activity data set; the user activity data set is analyzed to extract the key features of the user activity data set and construct a machine learning model of the user's motion pattern and frequency of participation in activities. The trained machine learning model is used to make real-time predictions on the newly collected user activity data to determine whether the user performs the corresponding sports project within the optimal exercise time, output the user's motion pattern and frequency of participation in activities, match the output results with the recommended content, and evaluate the effectiveness of the recommended content; the recommendation model is optimized and adjusted according to the analysis and output results of the user activity data.

[0043] For example, the GPS positioning data of the user's mobile phone is obtained through the application embedded SDK, such as Android's Google Location API or iOS's Core Location framework, to collect the user's location information, including latitude and longitude, altitude, speed, etc.; at the same time, the built-in accelerometer sensor of the mobile phone is used to record the user's motion acceleration data through Android's SensorManager class or iOS's Core Motion framework to obtain the user's motion trajectory and motion intensity information. Add the corresponding dependency library in the application's build.gradle or Podfile, and declare the corresponding permissions in the application's AndroidManifest.xml or Info.plist, and then call the corresponding API interface in the application's Java or Objective-C code, such as requestLocationUpdates() or startAccelerometerUpdates(), to start data collection. Set up a behavior log collection module in the application to record various operations of the user in the application, such as browsing recommended content, clicking to accept recommendations, starting exercise, ending exercise, etc., to obtain the interaction data between the user and the recommended content, and form a structured behavior log database. The collected GPS positioning data is cleaned and preprocessed to remove abnormal values ​​(such as sudden position changes, speeds beyond the normal range, etc.) and invalid data (such as low positioning accuracy, insufficient number of satellites, etc.), and the GPS data is smoothed using the Kalman filter algorithm to eliminate noise interference and improve positioning accuracy. The specific implementation steps include: defining system state variables (such as position, speed) and observation variables (such as GPS coordinates), establishing a state space model; initializing the prior estimates and covariance matrix of state variables; looping through the prediction step (predicting the current state estimate and covariance matrix based on the state estimate and state transfer matrix at the previous moment) and the update step (calculating the Kalman gain based on the current observation and observation matrix, updating the state estimate and covariance matrix), and outputting the filtered state estimate, such as the smoothed position coordinates. The accelerometer data is normalized to remove the influence of gravity acceleration and extract the user's motion acceleration characteristics. The pre-processed GPS positioning data and accelerometer data are integrated with the behavior log data, and these data are synchronized and aligned according to the timestamp to construct a multi-dimensional user activity data set. Each data contains fields such as user ID, timestamp, location information, motion acceleration, behavior type, etc. The user activity data set is analyzed using machine learning algorithms, and algorithms suitable for processing time series data and high-dimensional data are selected, such as random forests, recurrent neural networks (RNNs), etc.Through feature engineering, key features in the data set are extracted, such as time domain features (such as mean, variance, kurtosis, etc.), frequency domain features (such as spectral energy, spectral entropy, etc.), time-frequency domain features (such as statistical features of wavelet coefficients, etc.), and feature selection methods such as filtering, wrapping, embedding, etc., such as information gain, recursive feature elimination, L1 regularization, etc., are used to select the most relevant or important feature subsets, and train the prediction model of user movement patterns and frequency of participation in activities. Use the trained machine learning model to make real-time predictions on newly collected user activity data. To ensure real-time performance and performance, the following optimization measures can be adopted: Use model pruning, quantization, distillation and other technologies to reduce the size and amount of calculation of the model and speed up the prediction speed; use hardware accelerators such as GPU and TPU, or use acceleration libraries such as TensorRT and NNAPI to improve the parallel computing capability of the model; use stream processing frameworks such as Kafka and Flink to perform real-time cleaning, aggregation, feature extraction and other operations on data to reduce the data preparation time during prediction; use cache systems such as Redis and Memcached to cache hot data or intermediate results to avoid repeated calculations. Finally, it determines whether the user has performed the exercise according to the recommendation, outputs the user's exercise mode (such as walking, running, cycling, etc.) and the frequency of participating in the activity (such as 3 times a week, 5 times a month, etc.), matches the prediction results with the recommended content, and evaluates the effectiveness of the recommendation. According to the analysis and prediction results of user activity data, the recommendation algorithm is optimized and adjusted. For example, according to the user's actual exercise mode and frequency, the recommended exercise type and frequency are adjusted to improve the personalization and accuracy of the recommendation; according to the user's acceptance and execution of the recommended content, the recommended time strategy and push method are adjusted to improve user participation and satisfaction. When the SDK is embedded in the application, Android's Google Location API can be used to create a LocationRequest object, set the positioning frequency to once per second, the positioning accuracy to GPS_PROVIDER, and then call the requestLocationUpdates() method of FusedLocationProviderClient to start location updates. For accelerometer data, you can use Android's SensorManager class, getDefaultSensor(Sensor.TYPE_ACCELEROMETER) to get the accelerometer sensor instance, then call the registerListener() method to register the sensor listener and set the sensor sampling frequency to 50Hz. When cleaning data, you can set the GPS positioning accuracy threshold to 10 meters and the satellite number threshold to 4, and remove data points that exceed the threshold.In the Kalman filter algorithm, the state transfer matrix F can be set to [[1,△t],[0,1]], the observation matrix H can be set to [[1,0],[0,1]], the state noise covariance matrix Q can be set to a diagonal matrix, the observation noise covariance matrix R can be set to a diagonal matrix, the initial state estimate can be the first valid GPS point, and the initial covariance matrix can be set to the unit matrix. In feature engineering, the mean, standard deviation, maximum, minimum and other time domain features of the user's motion speed can be extracted, the spectrum energy and spectrum entropy of the user's motion acceleration can be extracted, and the mean and variance of the wavelet coefficients of the user's motion trajectory can be extracted. In feature selection, the recursive feature elimination method can be used, the number of candidate features can be set to 50, the performance score of the feature subset can be calculated through cross-validation, and the feature with the lowest score can be iteratively eliminated until the target number of features is reached. In model training, the random forest algorithm can be selected, the number of decision trees can be set to 100, the maximum tree depth can be set to 10, the hyperparameters can be optimized using grid search, the model performance can be evaluated through 5-fold cross-validation, and the model with the highest F1 value can be selected as the optimal model. When making real-time predictions, you can use TensorFlowLite to convert the trained model into a mobile-optimized model, set the number of threads to 4, the batch size to 64, and use Android's NNAPI acceleration framework to achieve a single prediction latency of less than 10 milliseconds on a Qualcomm Snapdragon 855 processor. When optimizing the recommendation algorithm, you can personalize the recommended exercise type and frequency based on the user's exercise mode and frequency. For example, for users who like walking, you can appropriately increase the recommendation frequency, and for users who like cycling, you can appropriately extend the recommendation time. At the same time, dynamically adjust the recommendation time and push method based on the user's acceptance of the recommended content. For example, for users with high acceptance, you can appropriately increase the push frequency, and for users with low acceptance, you can try to change the push time and push channel.

[0044] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A campus life recording method integrating a large model, characterized in that: The method comprises: By collecting the school's curriculum data, including the class time and location information of each class, important dates in the school calendar, and the time schedule of school activities, the curriculum data is integrated and standardized to build a school-wide curriculum and activity database; Based on the school's curriculum data, determine the course availability information for each time period, and combine it with school activity data to determine the course availability and demand for activity venues during activity periods; Set up a prediction model based on time series prediction or regression analysis, input the school's curriculum schedule, school activities and historical venue usage data, and output the predicted value of venue usage rate in each time period by training and optimizing the prediction model parameters, and identify the expected peak period; Obtain the user's personal course schedule data, extract the user's free time data, and at the same time, analyze the user's exercise characteristics from the user's historical exercise data, including exercise type, exercise intensity and exercise habits, and establish a user exercise characteristic table; Based on the user's sports feature table and the venue's peak hours, collaborative filtering or content-based recommendation models are used to generate personalized sports time recommendations for each user, including obtaining the user's best sports time based on the user's free time, sports features, and venue peak hours; Regularly collect user feedback data, including user satisfaction with the optimal exercise time, the difference between the actual exercise time and the optimal exercise time, and update the user exercise feature table based on the feedback data; The above mentioned curriculum timetable data of the whole school is collected, including the class time and location information of each class, important dates in the school calendar, and the time arrangement of school activities, and the curriculum timetable data is integrated and standardized to build a curriculum and activity database for the whole school, including: Obtaining the curriculum data of all classes in the school, and performing structured processing and standardization on the curriculum data to form a curriculum data set; Analyze the school calendar, extract the exact start and end dates of the semester and the start and end time range of the exam week and the holiday schedule, determine whether each day is a normal school day through time overlap and intersection calculation, and associate and match the determination result with the course schedule data set; Extract the time intervals of various seasonal activities of the school, compare the activity time intervals with the course schedule data set, identify the temporary courses that need to be adjusted or suspended during the activities, and mark the temporary courses; Build a data model combining the school calendar and class schedule based on the collected time, location, courses and classes; The method of determining the free time of each time period based on the school's curriculum data and judging the free time of each time period and the demand for the use of activity venues during the activity period in combination with the school's activity data includes: By analyzing the course schedule data, counting the number and distribution information of courses in each time period, calculating the course occupancy rate of each time period, determining the course idleness in different time periods according to the course occupancy rate, and obtaining a time occupancy heat map of the course idleness information of the whole school; Obtain important event schedules in the school calendar, extract the time, location and type of the event schedule, and determine the demand for the use of the event venue; Overlay and analyze the time information of the school activity data with the time occupancy heat map, calculate the intensity index of venue resources in each time period during the activity, and identify the peak time period of the activity and the hot time period of each venue; By statistically analyzing the classroom information in the school's curriculum data, the occupancy information of each classroom in different time periods is obtained, and the idle classrooms in each time period are analyzed to determine whether the idle classrooms meet the use requirements of the activity venue; Acquire the reservation and use information of sports venues, combine the course schedule data and the school activity data, obtain the venue occupancy rate and free space in each time period, and generate venue scheduling suggestions to ensure that the free time period meets the use demand of the activity venue under the premise of ensuring physical education teaching; The method generates personalized exercise time recommendations for each user based on the user's exercise feature table and the venue peak period by using collaborative filtering or a content-based recommendation model, including obtaining the user's optimal exercise time based on the user's free time, exercise features and venue peak period, including: Extract the sports preference, sports intensity and sports time of each user's sports characteristics from the user sports feature table, obtain the congestion level of different sports venues in different time periods from the peak period data table, and construct a user-time period-venue triple data structure; Calculate the similarity of exercise preferences between different users, determine the k neighbor users with the highest similarity to the target user's exercise preferences, and select the exercise periods of the k neighbor users as candidate exercise periods; Analyze the project features of each sport, including the sport type, sport intensity, and required venue, build a knowledge graph of the sport, form a semantic feature vector, and calculate the similarity between the user feature vector and the sport feature vector; Based on the user's free time, exercise preference and venue congestion, the candidate exercise time periods are scored and sorted, and the candidate exercise time period with the highest score is selected as the best exercise time; It also includes: if the user's free time coincides with the peak period of the venue, adjusting the recommended optimal exercise time according to the user's exercise intensity and exercise duration; if the user's free time is greater than a threshold, making the recommended optimal exercise time not adjacent to the peak period of the venue; If the user's free time coincides with the peak period of the venue, the recommended optimal exercise time is adjusted according to the user's exercise intensity and exercise duration; if the user's free time is greater than a threshold, the recommended optimal exercise time is not adjacent to the peak period of the venue, including: Obtain the user's idle time data and venue peak time data, convert the user's idle time and venue peak time into time segments, and calculate the overlap of the time segments; If the user's free time coincides with the venue's peak period, the user's exercise intensity and exercise duration are extracted, and the exercise intensity interval and exercise duration interval are calculated based on the exercise intensity and exercise duration; During the user's free time, the time period adjacent to the site's peak period and with a pressure index below the threshold is selected as a candidate recommendation period; According to the user's exercise intensity range and exercise duration range, the candidate recommended time periods are screened, the similarity is calculated, and the candidate recommended time period with the highest similarity is selected as the optimal exercise time; If the user's free time is greater than a threshold, the best exercise time is selected from the off-peak period, and the difference between the off-peak period and the venue's peak time is greater than a preset threshold; The user feedback data is collected regularly, including the user's satisfaction with the optimal exercise time, the difference between the actual exercise time and the optimal exercise time, and the user exercise feature table is updated according to the feedback data, including: Acquire the feedback data, perform data cleaning, deduplication and formatting on the feedback data, extract key fields, and construct a structured feedback data set; Performing statistical analysis on the feedback data set, calculating average satisfaction scores in different time periods, drawing a satisfaction distribution histogram, and determining time periods in which the average satisfaction score is below a threshold; and calculating the average deviation and deviation distribution of the actual movement time and the optimal movement time, and determining the time period in which the average deviation is higher than a threshold; For the time period when the average satisfaction score is lower than the threshold and the average deviation is higher than the threshold, key features are identified through feature selection and correlation analysis, and the weights and thresholds of the key features are adjusted to modify the parameters of the recommendation model; Dynamically adjust the parameters of the recommendation model based on the user's satisfaction score for the optimal exercise time; Regularly update the actual exercise time and satisfaction score in the user's feedback data to the user's exercise feature table to replace or supplement the original exercise feature value; It also includes: regularly collecting users' satisfaction with the optimal exercise time and the deviation data between the actual exercise time and the optimal exercise time through online questionnaires and in-app feedback functions; combining the activity data recorded by the user's terminal and the behavior log in the application to evaluate whether the user exercises according to the recommended content, the recommended content includes the optimal exercise time and the corresponding sports events, and identifying the user's exercise pattern and the frequency of participation in activities.

2. The method according to claim 1, characterized in that The prediction model is set based on time series prediction or regression analysis, and the school's course schedule, school activities and historical venue usage data are input. By training and optimizing the prediction model parameters, the predicted values ​​of venue usage rate in each time period are output, and the expected peak period is identified, including: Collect and organize the school's course schedule, school activities and historical venue usage data, clean, standardize and format the data, extract key features related to venue usage rate through feature engineering, including course type, activity type and time period, and build a structured training data set; Set up prediction models based on time series prediction or regression analysis, including long short-term memory neural networks or time convolutional networks, and determine the input and output formats, network structures, and hyperparameters of the prediction models according to the characteristics of the training data set and the prediction target; Tune hyperparameters using grid search or Bayesian optimization, including learning rate, batch size, number of hidden layers, and number of neurons; Optimize the parameters of the prediction model through cross-validation.

3. The method according to claim 1, characterized in that The user's personal course schedule data is obtained and the user's free time data is extracted; at the same time, the user's exercise characteristics, including exercise type, exercise intensity and exercise habits, are analyzed from the user's historical exercise data to establish a user exercise characteristic table, including: By connecting with the school's education system, the user's personal course schedule data is obtained, and the course schedule, including the course name, class time, and course duration, is extracted. Based on the course schedule, the user's free time data is determined to form the user's free time table; Obtain the types of sports that the user has participated in from the user's historical sports records, including basketball, football and swimming, count the number of times and duration of participation in each type of sports, obtain the user's preference for different types of sports, analyze the frequency and duration distribution of the user's participation in sports in different time periods, and determine the user's exercise time habits; Estimate the exercise intensity of the user in each exercise according to the duration and frequency of the user's participation in different types of exercise, calculate the average speed by running distance and time, and comprehensively evaluate the user's consumption and exercise intensity level in different sports according to the duration and intensity classification of ball sports to obtain the user's exercise intensity characteristics; Constructing the user exercise feature table according to the user's exercise type preference, exercise time habit and exercise intensity characteristics; Analyze the user's free timetable and exercise time habits, determine the time period when the user is accustomed to exercise in his free time, and preliminarily screen out the user's sports items of interest and preferred exercise time periods in combination with the user's sports type preferences; According to the user's exercise intensity characteristics and physical health information, the sports of interest are refined and adjusted. Combined with the learned venue information, the user's exercise time period and exercise type are matched, the recommended sports venue corresponding to the user is determined, and an exercise recommendation plan is generated.

4. The method according to claim 1, characterized in that: The user's satisfaction with the optimal exercise time and the deviation data between the actual exercise time and the optimal exercise time are collected regularly through online questionnaires and in-app feedback functions, including: Regularly collect user feedback texts through online questionnaires and in-app feedback functions, and identify the satisfaction with the optimal exercise time and the deviation data between the actual exercise time and the optimal exercise time in the feedback texts; Performing natural language processing on the feedback text, including word segmentation, stop word removal, and part-of-speech tagging; Performing semantic analysis and sentiment analysis on the feedback text, and identifying key topics and sentiment tendencies in the feedback text through word frequency statistics or TF-IDF weight calculation; Conduct descriptive statistical analysis on quantitative data, calculate the mean, median, and standard deviation of satisfaction scores, and draw a satisfaction distribution histogram; Calculate the absolute deviation and relative deviation of the deviation data between the actual exercise time and the optimal exercise time, and analyze the distribution characteristics of the deviation; According to the analysis results of the user satisfaction and deviation data, the time period when the satisfaction is lower than the threshold or the deviation is higher than the threshold is identified, and combined with the feedback text, the key factors affecting the user satisfaction and deviation are identified.

5. The method according to claim 1, characterized in that The method combines the activity data recorded by the user terminal and the behavior log in the application to evaluate whether the user exercises according to the recommended content, the recommended content includes the best exercise time and the corresponding sports items, and identifies the user's exercise pattern and the frequency of participating in activities, including: The embedded SDK of the application obtains the GPS positioning data of the user's mobile phone and collects the user's location information; the built-in accelerometer sensor of the terminal is used to record the user's motion acceleration data and obtain the user's motion trajectory and motion intensity information; a behavior log collection module is set in the application to obtain the interaction data between the user and the recommended content and form a structured behavior log database; the collected GPS positioning data is cleaned and preprocessed, and the GPS positioning data is smoothed using the Kalman filter algorithm; the motion acceleration data is normalized to extract the user's motion acceleration characteristics; the preprocessed GPS positioning data and the motion acceleration data are fused with the behavior log data to construct a user activity data set; the user activity data set is analyzed to extract the key features of the user activity data set, and a machine learning model of the user's motion pattern and frequency of participation in activities is constructed. The trained machine learning model is used to make real-time predictions on the newly collected user activity data, determine whether the user performs the corresponding sports project within the optimal exercise time, output the user's motion pattern and frequency of participation in activities, match the output results with the recommended content, and evaluate the effectiveness of the recommended content; the recommendation model is optimized and adjusted according to the analysis and output results of the user activity data.

Images