Method for collecting service operation information in network information service system

Abstract

The invention discloses a method for collecting service operation information in a network information service system, which comprises the following steps: S1, a user terminal is provided with a local module, and an operation sequence buffer area stores recent operation as required and updates and clears abnormities according to first-in first-out; s2, arranging data in a buffer area, extracting features, cleaning anomalies, judging a behavior pattern by using sequence pattern mining in combination with a threshold value and a machine learning algorithm, and updating the model regularly; s3, constructing an evaluation model for different behavior modes, dynamically setting a weight by integrating multiple factors, and performing regular re-evaluation adjustment; s4, scores are calculated according to behavior modes and weights, and a verification mechanism is set to be compared with historical data to ensure accuracy; and S5, comparing the score with a threshold value to determine a collection granularity strategy, increasing and decreasing information details, and ensuring high efficiency and stability of terminal resources. According to the method, information collection is accurately adjusted for user behaviors, resource utilization, transmission and server performance are improved, personalization and stability are optimized, user experience is improved, and network information service upgrading is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of information collection and processing, and in particular to a method for collecting service operation information in a network information service system. Background Art

[0002] With the widespread popularity of online information services in mobile news reading, live video social platforms, and enterprise-level collaborative office software, the amount of data has exploded. However, traditional online information collection strategies have serious drawbacks.

[0003] Traditional strategies mostly adopt fixed modes, ignoring differences in user behavior while collecting information. This leads to a series of problems in various application scenarios: on terminal devices, a large amount of redundant data is stored, memory and CPU resources are over-consumed, causing device operation to be stuck, energy consumption to increase, battery life to shorten, and it is difficult to push personalized services; at the network transmission level, it is impossible to dynamically adjust according to network conditions and user behavior, causing data transmission delays, stuck or even lost, and single-channel transmission cannot fully utilize network resources, reducing transmission efficiency and success rate; on the server side, it is necessary to process massive amounts of unfiltered raw data, the CPU load is too high, and the memory usage is too much. It is easy to have slow responses or even downtime at high concurrency, which seriously hinders the efficient operation of business processes and the improvement of service quality. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method for collecting service operation information in a network information service system.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for collecting service operation information in a network information service system comprises the following steps:

[0007] S1. Operation information collection and caching: A local information collection module is set up in the user terminal, and its operation sequence buffer records the user's recent operation information. The storage range is set according to the device and application requirements, and it is updated according to the first-in-first-out method, and can automatically clean up abnormal operation data;

[0008] S2. Behavior pattern analysis and determination: Preliminary screening and sorting of buffer data, extraction of key features and cleaning of abnormal data, calculation of support of operator subsequences using sequence pattern mining algorithms, determination of behavior patterns based on thresholds and machine learning classification algorithms, and regular updating and optimization of machine learning models based on new data;

[0009] S3. Importance assessment parameter setting: Construct an information importance assessment model for different behavior patterns, dynamically allocate operation importance weights based on multiple factors, and set up a dynamic weight adjustment mechanism to regularly re-evaluate and adjust the weights;

[0010] S4. Importance score calculation and verification: The score is calculated according to the behavior pattern and weight according to the relevant formula, and a verification mechanism is set up to compare with historical data to ensure that the score is accurate and reliable;

[0011] S5. Collection granularity decision and optimization: The collection granularity adjustment strategy is determined based on the comparison between the score and the threshold, increasing or decreasing the collection information details, and the resource status of the terminal device is considered during the adjustment to ensure efficient and stable operation.

[0012] Preferably, the operation sequence buffer uses an efficient data compression algorithm to reduce memory usage and improve read and write speeds, and automatically selects the optimal compression algorithm based on the data type and distribution characteristics of the operation information, such as Huffman coding compression for text and differential coding compression for numerical values.

[0013] Preferably, the machine learning classification algorithm adopts an integrated learning method to fuse the prediction results of multiple basic classification models, and formulates the final classification result through voting or weighted average mechanism to enhance accuracy and robustness and reduce the risk of misjudgment.

[0014] Preferably, in the importance weight dynamic configuration step, the weight adjustment period is adaptively set according to the frequency of application business changes and the stability of user behavior, shortened when the business is changeable or the user behavior fluctuates greatly, and extended when the business is stable or the user behavior is fixed.

[0015] Preferably, the sequential pattern mining algorithm adopts parallel computing technology when calculating support, divides the operation sequence data into blocks for simultaneous calculation by multiple computing cores or nodes, and aggregates the results to obtain the final support to improve pattern recognition efficiency, which is suitable for large-scale data processing.

[0016] Preferably, there are corresponding specific adjustment strategies in different application scenarios: in the mobile news reading application scenario, the behavior pattern and score are determined according to the news browsing related operation type to adjust the news picture data collection; in the video live social platform scenario, the behavior pattern and score are determined according to the live interactive operation type to adjust the interactive information collection and transmission; in the enterprise-level collaborative office software scenario, the behavior pattern and score are determined according to the document collaboration operation type to adjust the document information collection and preprocessing.

[0017] Preferably, in the collection granularity adjustment decision step, when the collection granularity is adjusted, a corresponding collection strategy adjustment log is generated at the same time, recording the reason and time of the adjustment, detailed change information of the collection granularity before and after, and the estimated impact on device resources and network transmission, so as to facilitate reference in subsequent data analysis, troubleshooting and strategy optimization.

[0018] The present invention has the following beneficial effects:

[0019] 1. Through adaptive collection strategies, we can accurately analyze user behavior patterns and optimize terminal resources. We can effectively reduce memory usage by about 30% and CPU usage by about 20%, reduce data redundancy, improve application response speed and battery life, and push news content that better suits users' interests, enhance the personalization of reading experience, and avoid the problems of terminal resource waste and insufficient personalization.

[0020] 2. Efficiently manage information collection and transmission with adaptive strategies based on user behavior and network conditions. By accurately identifying behavioral patterns and reducing non-critical data collection, data transmission volume has dropped by about 40%. Combined with the intelligent channel selection mechanism, the transmission success rate has increased by about 18% based on dynamic bandwidth allocation. This ensures clear and smooth live broadcast images and instant delivery of interactive information. Both the audience viewing experience and the interaction between the anchor and the audience are significantly enhanced, which effectively promotes user activity and business expansion of the live broadcast platform, and avoids the lack of flexibility in information collection and transmission, which can easily lead to network freezes and poor live broadcast experience.

[0021] 3. Relying on adaptive collection and local preprocessing strategies, according to the user's document collaboration behavior pattern, key data is screened and preprocessed, and the server processing time for the same task is significantly shortened by about 30%. The peak CPU utilization rate is reduced by about 25%, and the memory usage is reduced by about 20%, effectively alleviating the pressure on the server during high concurrency periods. Ensure that the internal document sharing, task collaboration and other processes of the enterprise are carried out efficiently and orderly, avoid business delays caused by server performance bottlenecks, and improve the overall office efficiency and competitiveness of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic flow chart of a method for collecting service operation information in a network information service system proposed by the present invention;

[0023] Figure 2 This is a code display diagram for collecting and caching operation information, managing user terminal operation data records, updating, and cleaning up exceptions in Example 1;

[0024] Figure 3 This is a code display diagram for behavior pattern analysis and determination and model update in Example 2;

[0025] Figure 4 This is a code display diagram for Example 3 involving dynamic configuration of importance weights, score calculation verification, collection granularity decision optimization, and implementation of specific adjustment strategies for different application scenarios. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0027] A method for collecting service operation information in a network information service system comprises the following steps:

[0028] S1. Operation information collection and caching: A local information collection module is set up on the user terminal, and its operation sequence buffer is responsible for recording the user's recent operation information. The storage range is set according to the specific needs of the device and application, and data updates follow the first-in-first-out principle, and can automatically detect and clean up abnormal operation data. This buffer uses an efficient data compression algorithm, such as Huffman coding compression for text-based operation information and differential coding compression for numerical types. The compression algorithm is automatically selected based on the data type and distribution characteristics of the operation information, thereby reducing memory usage and improving read and write speeds.

[0029] S2. Behavior pattern analysis and judgment: First, the buffer data is preliminarily screened and sorted, key features are accurately extracted, and abnormal data is effectively cleaned. The sequence pattern mining algorithm is used to calculate the support of the operation subsequence, and the behavior pattern is determined by combining the set threshold and the machine learning classification algorithm (using an integrated learning method, integrating the prediction results of multiple basic classification models, and using voting or weighted average mechanisms to determine the final classification results, enhancing accuracy and robustness, and reducing the risk of misjudgment). At the same time, the machine learning model will be regularly updated and optimized based on new data to ensure that the judgment of the behavior pattern can keep pace with the times and adapt to data changes.

[0030] S3. Importance evaluation parameter setting: A special information importance evaluation model is built for different behavior patterns, and a variety of factors are comprehensively considered to dynamically assign operation importance weights. The weight adjustment cycle is adaptively set based on the frequency of application business changes and the stability of user behavior. When the business is changeable or the user behavior fluctuates greatly, the cycle is shortened, and when the business is stable or the user behavior is fixed, the cycle is extended. In addition, a dynamic weight adjustment mechanism is set to regularly re-evaluate and adjust the weight to ensure the rationality and timeliness of the weight setting.

[0031] S4. Importance score calculation and verification: According to the determined behavior patterns and corresponding weights, the scores are accurately calculated according to the relevant formulas. At the same time, a verification mechanism is established to compare the calculated scores with historical data to ensure the accuracy and reliability of the scores, providing a solid data basis for subsequent decision-making.

[0032] S5. Decision and optimization of collection granularity: The adjustment strategy of collection granularity is determined by comparing the importance score with the set threshold, which can increase or decrease the details of collected information. During the adjustment process, the resource status of the terminal device is fully considered to ensure the efficient and stable operation of the system. When adjusting the collection granularity, the corresponding collection strategy adjustment log is generated synchronously, which records in detail the reasons, time, specific changes in the collection granularity before and after the adjustment, and the estimated impact on device resources and network transmission, providing a strong reference for subsequent data analysis, troubleshooting and strategy optimization. There are corresponding specific adjustment strategies in different application scenarios: in the mobile news reading application scenario, the behavior pattern and score are determined according to the type of news browsing related operations, and then the news picture data collection is adjusted in a targeted manner; in the video live social platform scenario, the behavior pattern and score are determined according to the type of live interactive operation, so as to flexibly adjust the interactive information collection and transmission; in the enterprise-level collaborative office software scenario, the behavior pattern and score are determined according to the document collaboration operation type, so as to reasonably adjust the document information collection and preprocessing. The sequential pattern mining algorithm uses parallel computing technology to calculate support, divides the operation sequence data into blocks and calculates them simultaneously on multiple computing cores or nodes, and finally summarizes the results to obtain the final support, effectively improving the efficiency of pattern recognition, which is especially suitable for large-scale data processing scenarios.

[0033] Embodiment 1:

[0034] Step 1: Operation information collection and caching:

[0035] The user behavior adaptive collection module is integrated into the mobile news reading application. The operation sequence buffer of the local information collection module is set to store the user's most recent 20 operation information, including news browsing (O1), searching news (O2), commenting on news (O3), sharing news (O4) and other operation types, and records the corresponding news category, keyword, operation time interval, application scenario (such as on the home page, channel page, etc.) and device status (such as network connection type, power, etc.). When a new operation is added and the number of records exceeds 20 times, the earliest operation record is automatically deleted. For example, if the user searches for news, browses news, shares news and other operations in sequence, the buffer will record these operation information in sequence. If the buffer is full and there is a new operation, such as searching for news again, the earliest search news operation record will be deleted. At the same time, the buffer has the function of automatically cleaning up abnormal operation data. For example, repeated search operation records caused by temporary network failures will be identified and cleared.

[0036] Step 2: Behavior pattern analysis and determination:

[0037] The buffer data is preliminarily screened to extract key features such as the number of searches for a specific news category and the distribution of browsing time for news, and to clean up abnormal data caused by network fluctuations, such as a large number of identical operation records in a very short period of time. Then, the sequence pattern mining algorithm is used to calculate the support of the operation subsequence. Assuming the total number of operation sequences is m, the number of operation sequences containing a specific subsequence is k, and the support formula is:

[0038]

[0039] The minimum support threshold is set to 0.2. For example, if the support of the subsequence "search news - browse multiple search results news" exceeds the threshold, it is determined that the user is in information search and browsing mode. Assuming that there are 100 operation sequences in a period of time, 30 of which contain the subsequence "search news - browse multiple search results news", then

[0040]

[0041] The model is identified by a threshold greater than 0.2. The machine learning classification algorithm uses an integrated learning method to integrate the prediction results of basic classification models such as decision trees and naive Bayes, and determines the final behavior pattern classification results through a weighted average mechanism. The machine learning model is updated and optimized every week based on the newly accumulated user operation data to adapt to the content updates of news reading applications and changes in user habits.

[0042] Step 3: Importance assessment parameter setting:

[0043] Construct an information importance assessment model to evaluate the importance weight of browsing news detail pages in information search and browsing mode Quickly switch the news list operation weight (Assume that the quick switching news list is O5.) The importance weight is set based on big data analysis of user reading habits and news application business needs, and a dynamic adjustment mechanism is established to re-evaluate and adjust monthly based on changes in user reading trends (such as changes in attention to specific news types).

[0044] Step 4: Importance score calculation and verification:

[0045] According to the formula:

[0046]

[0047] Calculate the importance score of the operation information. For example, if a user quickly switches the news list 3 times within 10 seconds (the average time interval between each switch operation is 3 seconds), and browses a news detail page for 2 seconds, then the importance score is:

[0048] I = 0.6*2+0.2*(3+3+3)=2.4

[0049] At the same time, this score will be compared and verified with the average score and standard deviation of the same type of historical behavior mode (such as the average score of information search and browsing mode in the past week). If the score deviates from the normal range (such as exceeding the average score ± standard deviation range), the operation data records, weight configuration and calculation process will be reviewed to ensure that the score is accurate and reliable.

[0050] Step 5: Collect granularity decision adjustment mechanism

[0051] When I < 5 (set threshold), it indicates that the user is in a fast browsing state. At this time, the collection of high-definition version data of news pictures is reduced, and only low-resolution thumbnail data is collected to reduce data storage. When adjusting the collection granularity, consider the current resource usage of the terminal device, such as CPU load, memory usage, network bandwidth utilization, etc. If device resources are tight (such as CPU usage higher than 80% or memory usage higher than 70%), further reduce the collection of non-critical information, such as reducing the collection of some metadata of news comments, to ensure that the application runs smoothly and the battery consumption rate is also slowed down.

[0052] Embodiment 2:

[0053] Different from the first embodiment, this embodiment further has the following contents:

[0054] Step 1: Operation information collection and caching:

[0055] The operation sequence buffer records the user's most recent 30 operations, including entering the live broadcast room (O1), sending barrage (O2), sending gifts (O3), switching live broadcast rooms (O4), etc. The operation objects are the live broadcast room ID, gift type, etc. The operation time interval is recorded together, and the application scenario during the operation (such as in the popular recommended live broadcast room list, follow the anchor's live broadcast room page, etc.) and the device status (such as device temperature, network delay, etc.) are also recorded. The buffer size is 30 operation records, updated on a first-in-first-out basis, and can automatically identify and clean up abnormal operations, such as rapid and continuous entry and exit of live broadcast room operation records caused by accidental touches. For example, a series of operations performed by a user during a live broadcast will be recorded in sequence. When the number of records reaches 30, the earliest operation record will be replaced by the new operation to ensure that the buffer data reflects the user's recent behavior.

[0056] Step 2: Behavior pattern analysis and determination:

[0057] First, extract key features from the buffer data, such as the frequency of entering a specific live broadcast room, the time interval distribution of sending bullet comments, etc., and clean up abnormal data, such as repeated bullet comment operation records caused by network jams. Use the sequence pattern mining algorithm to calculate the support of the operation subsequence, the formula is:

[0058]

[0059] The minimum support threshold is set to 0.25. For example, if the support of the subsequence "enter the live broadcast room - send bullet comments - continue to watch for a period of time" reaches the threshold, it is recognized that the user is in the mode of deep participation in the live broadcast room. Assuming that there are 200 operation sequences in total, 60 of which contain this subsequence, then:

[0060]

[0061] Among them, the machine learning classification algorithm adopts an integrated learning method, integrating the results of decision trees, support vector machines and other models, determines the final classification through a voting mechanism, and is updated and optimized based on new data every month to adapt to the functional updates of the live broadcast platform and changes in user interaction habits.

[0062] Step 3: Setting the importance assessment parameters

[0063] Build an information importance assessment model to send bullet screen operation weights in the deep participation live broadcast room mode Gift weight These weights are set based on the live streaming platform's business objectives (such as promoting interaction and increasing revenue) and user behavior analysis, and a dynamic adjustment mechanism is established to re-evaluate and adjust them every two weeks based on platform activities (such as during live streaming events with specific themes) and changes in user group behavior (such as changes in the style of barrage sent by a certain type of user).

[0064] Step 4: Importance score calculation and verification

[0065] According to the formula:

[0066]

[0067] Calculate the importance score. For example, if a user sends 3 bullet comments within 15 seconds (with an average interval of 4 seconds between each posting) and sends a gift once, and the gift sending operation takes 2 seconds, then the importance score is:

[0068] I = 0.7*(4+4+4)+0.8*2=10.4

[0069] At the same time, the score statistics are compared and verified with the historical score statistics under the same type of behavior mode. If there are any abnormalities, the data processing flow is rechecked to ensure the accuracy of the score.

[0070] Step 5: Collection granularity decision and transmission optimization mechanism

[0071] When I>8 (set threshold), increase the collection of user interaction details in the live broadcast room, such as the frequency of using emoticons to send bullet screens, the interactive response time to the anchor's specific topic, etc., and transmit this information after local compression processing. Use efficient data compression algorithms, such as Huffman coding compression for text-based bullet screen information, and differential coding compression for numerical gift amounts and other information to reduce the amount of data. At the same time, establish an intelligent transmission channel selection mechanism with the server. When the network detection module finds that the current network bandwidth is sufficient (such as bandwidth utilization is less than 60%), it will give priority to using high-speed channels to transmit important and uncompressed interactive information (such as large-amount transaction information for sending gifts); when the network is congested (bandwidth utilization is higher than 80%), the compressed interactive information will be transmitted through a backup channel to ensure the timeliness and integrity of data transmission. When adjusting the collection granularity, consider the device resource status. For example, when the device temperature is too high, appropriately reduce the collection of some non-critical interactive information to avoid device performance degradation affecting the live broadcast experience.

[0072] Embodiment three:

[0073] Compared with the first and second embodiments, in this embodiment:

[0074] Step 1: Operation information collection and caching:

[0075] The operation sequence buffer records the user's most recent 50 operations. The operation types include document editing (O1), file sharing (O2), task assignment (O3), and meeting participation (O4). The operation objects include document name, shared object, task details, etc. The operation time interval is recorded, and the project scenario (such as project development stage, project review stage) and device status (such as the stability of the device's office network connection, remaining storage space, etc.) in which the operation is performed are also recorded. The buffer is updated according to the first-in-first-out principle. After the storage capacity reaches 50 operations, the earliest record is automatically replaced, and abnormal operation data can be automatically cleared, such as repeated task assignment operation records caused by software failures. For example, the operations of employees during project collaboration will be recorded in the buffer in an orderly manner. If the buffer is full and there is a new operation, the earliest operation record will be deleted to ensure data timeliness.

[0076] Step 2: Behavior pattern analysis and determination:

[0077] Extract key features from the buffer data, such as the total duration of document editing, the frequency of file sharing, etc., and clean up abnormal data, such as repeated file sharing operation records caused by network fluctuations. Use the sequence pattern mining algorithm to calculate the support of the operation subsequence, the formula is:

[0078]

[0079] The minimum support threshold is set to 0.3. For example, if the support of the subsequence "document editing - multiple saving - file sharing" meets the threshold, it is recognized that the user is in the document collaboration and sharing mode. Assuming that a total of 300 operation sequences are analyzed, 120 of which contain this subsequence, then:

[0080]

[0081] Among them, the machine learning classification algorithm adopts an integrated learning method, integrating the results of multiple basic classification models, and determines the final classification through a weighted average mechanism. It is updated and optimized every quarter based on a large amount of newly accumulated enterprise employee operation data to adapt to enterprise business process adjustments and changes in employee work habits.

[0082] Step 3: Setting the importance assessment parameters

[0083] Build an information importance assessment model to evaluate the weight of key content modification operations in document editing in document collaboration mode File share operation weight The importance weights are set based on the company's management needs for document collaboration and data analysis results, and a dynamic adjustment mechanism is established to re-evaluate and adjust them every six months based on adjustments to the company's business strategy (such as changes in the direction of key projects) and changes in employee collaboration patterns (such as changes in team collaboration styles).

[0084] Step 4: Importance score calculation and verification

[0085] According to the formula:

[0086]

[0087] Calculate the importance score. For example, if a user has three key content modification operations during document editing, each operation takes 5 seconds, 4 seconds, and 3 seconds, and then performs a file sharing operation, which takes 2 seconds, then the importance score is:

[0088] I = 0.8*(5+4+3)+0.6*2=10.8

[0089] Compare and verify the scores with historical statistical data of scores under similar behavior patterns. If any anomalies are found, review the data processing flow to ensure the reliability of the scores.

[0090] Step 5: Collection granularity decision and service period linkage mechanism

[0091] When I>10 (set threshold), the historical version difference information of document modifications is collected in detail. For example, the key content of document modifications, such as newly added chapters and paragraphs, modified important data, etc., is extracted through text comparison algorithms and organized into a concise difference report. Before uploading to the server, the data is pre-processed locally to extract key modification points and difference data to reduce the amount of data. After the server receives the data, since the data has been optimized and filtered locally, when performing data storage and analysis, such as team collaboration efficiency analysis and document flow monitoring, the processing speed is increased by about 30% compared to traditional collection methods, the server CPU utilization rate is reduced by about 25%, and the memory usage is reduced by about 20%. When adjusting the collection granularity, the terminal device resource status is considered. For example, when the remaining storage space of the device is insufficient, data that has less impact on server-side analysis is compressed first to ensure the efficient and stable operation of the enterprise collaborative office process.

[0092] It should be noted that in the comparative example, in the mobile news application reading, 500 active users were tested in the case of a test period of one week. The information collection strategy adopted in the comparative example 1 is a traditional method of uniformly collecting all information without behavioral pattern analysis. The information collection strategy adopted in the comparative example 2 is a method of collecting information at fixed time intervals using a simple time threshold without considering the behavioral pattern. The example 1 is compared, as shown in Table 1:

[0093] Table 1: Advantages of this method in terminal resource optimization

[0094] Serial number Average terminal memory usage Average terminal CPU usage Data collection amount Embodiment 1 210MB 20％ 50MB Comparative Example 1 320MB 26％ 100MB Comparative Example 2 280MB 24％ 85MB

[0095] As shown in Table 1, in the mobile news reading application, Example 1 analyzes 20 user operations, mines the behavior pattern with a threshold of 0.2 in a sequence mode, and flexibly adjusts the collection granularity according to the weighted score, such as reducing the collection of high-definition pictures when the score is low. After a week of testing by 500 users, the memory usage was greatly reduced by about 30%, and the CPU usage rate was significantly reduced by about 20%, which effectively reduced the pressure on terminal resources, made news reading smoother, and the battery life longer, fully demonstrating the advantages of this method in optimizing resources according to user behavior. Comparative Example 1 adopts traditional fixed collection, and collects all data without considering behavioral differences, resulting in an increase in memory instead of a decrease, no improvement in CPU usage, and resources occupied by a large amount of redundant data. The device is prone to freeze and heat, which seriously affects the reading experience. Comparative Example 2 is based on simple time threshold collection. Although there is a certain resource optimization, it lacks accuracy. The memory is only reduced by 10% to 15%, and the CPU usage rate is reduced by 5% to 10%. It still collects a lot of useless data, and it is impossible to efficiently use resources, and it is difficult to meet the requirements of mobile reading for fluency. It can be seen that this method can accurately adapt to user behavior and dynamically adjust the collection strategy, which far exceeds traditional and simple strategies in terminal resource optimization, and effectively improves the performance and user experience of mobile news reading applications.

[0096] It should be noted that in the comparative embodiment, on the video live broadcast dating platform, under the condition of a test period of one month and one thousand live broadcast activities, the transmission strategy adopted by the comparative example 1 is no behavior pattern analysis method, fixedly collects all data of operations such as entering the live broadcast room, sending barrage, sending gifts, switching live broadcast rooms, etc., and always uses a single network channel for transmission. The transmission strategy adopted by the comparative example 2 is to give priority to the transmission of high-frequency operation information such as sending barrage and sending gifts, but does not perform data compression and comprehensive judgment of network conditions, and has no intelligent channel selection. The comparative example 2 is compared, as shown in Table 2:

[0097] Table 2: Advantages of this method in terms of network transmission efficiency

[0098] Serial number Total amount of transmission Transmission success rate High channel usage Backup channel usage Embodiment 2 500MB 95％ 40％ 20％ Comparative Example 1 850MB 80％ 0％ 0％ Comparative Example 2 725MB 82％ 0％ 0％

[0099] As shown in Table 2, in the video dating platform, Example 2 adaptively collects and transmits based on user behavior patterns and network conditions. By analyzing 30 operations, the sequence pattern mining sets a threshold of 0.25 to identify the deep participation in the live broadcast room mode, and adjusts the collection and transmission channels according to the weight calculation score. When the network bandwidth is low, the backup channel is used to transmit compressed information, and when it is high, the high-speed channel is used to transmit important uncompressed information. After a large number of live broadcast activities were tested, the data transmission volume was reduced by about 40%, and the transmission success rate was increased by about 18%, which effectively alleviated the network pressure and ensured the real-time and smooth live interaction, highlighting the advantages of this method in intelligently allocating resources and adapting to network changes. Comparative Example 1 adopts traditional fixed collection and single-channel transmission, does not analyze the behavior pattern, and uniformly collects and transmits all data through a fixed channel. There is no obvious reduction in the data transmission volume, and the transmission success rate is only 80%. It cannot be flexibly responded to when the network is congested, resulting in freezes, delays, and even information loss in the live broadcast screen, which greatly damages the user's viewing and interactive experience. Comparative Example 2 gives priority to the transmission of high-frequency operation information, but does not comprehensively consider network and data compression. The amount of data transmission is reduced slightly, about 10% to 20%, and the transmission success rate is limited, about 5% to 10%. When the network is not good, it is difficult to ensure efficient and stable information transmission, and it cannot meet the live broadcast platform's demand for high-quality services. It can be seen that this method can dynamically optimize information collection and transmission based on user behavior and network environment. Compared with traditional and simple strategies, it significantly improves network transmission efficiency and stability, and can provide a better service experience for video live social platforms.

[0100] It should be noted that in the comparative embodiment, in the enterprise-level office software, 20 teams are selected, each team has 20 people, and the test period is three months. The comparative example 1 adopts a server linkage strategy without behavior pattern analysis, collects all data of document editing, file sharing, task allocation, meeting participation and other operations, and then directly uploads them to the server. The comparative example 2 adopts a server linkage strategy of uploading unprocessed operation data in batches every 2 hours, without considering user behavior patterns and data optimization, and is compared with the embodiment 3, as shown in Table 3:

[0101] Table 3: Advantages of this method in terms of service period performance improvement

[0102] Serial number Processing time CPU Utilization Memory usage Local data preprocessing time Embodiment 2 7min 60％ 4GB 0.5 / S Comparative Example 1 10min 80％ 5GB 0 / S Comparative Example 2 9min 73％ 4.7GB 0 / S

[0103] As shown in Table 3, Example 3 is based on adaptive collection and local preprocessing of user behavior patterns. 50 operations are analyzed, and the sequence pattern mining sets a threshold of 0.3 to determine the document collaboration mode. The collection and local processing are adjusted according to the weight calculation score. After three months of testing by multiple enterprise teams, the server processing time for the same task is shortened by about 30%, the CPU utilization peak is reduced by about 25%, and the memory usage is reduced by about 20%. This strategy effectively reduces the burden on the server, improves the processing speed and stability, and fully demonstrates the advantages of this method in accurately adapting to user behavior to optimize server performance. Comparative Example 1 adopts traditional collection and direct upload without analyzing the behavior pattern, and all data is directly transmitted to the server without screening. The server processing time is long without obvious improvement, the CPU utilization is high, the memory usage is high, and it is easy to respond slowly or even crash during business peaks, which seriously hinders the smooth progress of the enterprise collaborative office process. Comparative Example 2 does not optimize the data in batches during scheduled upload. Although there is a timing mechanism, it is not optimized for behavior. The server performance is slightly improved, the processing time is only shortened by about 5% to 10%, the CPU utilization is reduced by about 5% to 15%, and the memory usage is reduced by about 5% to 10%, which cannot meet the needs of efficient data processing and business expansion of enterprises. It can be seen that this method can dynamically adjust information collection according to user behavior, significantly improve server performance, far exceeding traditional and simple strategies, and effectively ensure the efficient and stable operation of enterprise-level collaborative office software.

[0104] Specifically, in each embodiment, key code examples and corresponding analysis are written in Python language.

[0105] Furthermore, if Figure 1As shown, in Example 1, the operation sequence buffer is simulated through the operation_buffer list, and the add_operation_to_buffer function implements the function of adding operation records to the buffer and controlling the buffer size according to the first-in-first-out principle, ensuring that the most recent 20 operation records are always stored. This helps to focus on analyzing the user's recent behavior and avoid excessive memory occupation by data, reflecting the preliminary optimization of terminal memory resources. The operation_weights dictionary defines the importance weights corresponding to different operation types, and the calculate_importance_score function calculates the importance score based on the operation records in the buffer and the corresponding weights. This weight-based calculation method can be flexibly adjusted according to business needs, allowing the system to distinguish the importance of different operations for analyzing user behavior, and then provide a basis for subsequent decisions on collection granularity, so as to dynamically optimize data collection according to user behavior, avoid collecting too much useless data and increase the burden on terminal resources, and decide whether to reduce or collect data normally based on the calculated importance score and the set threshold (5 in the example here). For example, when the score is low, reduce the collection of large-capacity data such as high-definition pictures. This achieves the purpose of prioritizing key data collection, improving application operation efficiency and saving resources when terminal resources are limited, and fully demonstrates the advantages of adaptive collection strategies in terminal resource optimization.

[0106] Furthermore, if Figure 3 As shown, in Example 2, the add_op function is used to implement basic management of the operation buffer, ensuring that only the most recent 30 operation type records are retained to facilitate focusing on recent behaviors. The check_behavior function determines the behavior pattern by simply counting the number of times the operation type occurs. Although it simplifies the logic, it reflects the idea of ​​mining the degree of user participation from the operation, preparing for subsequent policy adjustments, and the calc_score function quickly calculates the importance score based on the set weight to measure the current behavior value. The adjust function combines the behavior pattern, score, and simulated network bandwidth to decide whether to use a high-speed channel, a backup channel, or conventional processing, demonstrating the core logic of adaptively adjusting the collection and transmission strategy according to different situations, thereby optimizing network transmission.

[0107] Furthermore, if Figure 4As shown, in Example 3, the add_doc_op function is used to manage the operation buffer, store the operation information in the form of a dictionary and control it within 50 times, so as to facilitate the subsequent analysis of relevant operation details. The check_behavior function simply determines whether it is in the key mode of document collaboration based on the order of operations, and can screen out the behavior stages that have an important impact on server data processing, and avoid invalid data interfering with server performance analysis. The calc_score function calculates the operation importance score according to the set weight, and reasonably distinguishes the contribution of different operations to the focus of the server. The preprocess function performs a simple local preprocessing example on document-related operations in the key behavior mode, such as extracting some key difference content, aiming to reduce the amount of data uploaded to the server and optimize the data quality, thereby improving server performance, highlighting the advantages of this adaptive strategy in this regard.

[0108] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for collecting service operation information in a network information service system, characterized in that: The following steps are involved: S1. Operation information collection and caching: A local information collection module is set up in the user terminal, and its operation sequence buffer records the user's recent operation information. The storage range is set according to the device and application requirements, and it is updated according to the first-in-first-out method, and can automatically clean up abnormal operation data; S2. Behavior pattern analysis and determination: Preliminary screening and sorting of buffer data, extraction of key features and cleaning of abnormal data, calculation of support of operator subsequences using sequence pattern mining algorithms, determination of behavior patterns based on thresholds and machine learning classification algorithms, and regular updating and optimization of machine learning models based on new data; S3. Importance assessment parameter setting: Construct an information importance assessment model for different behavior patterns, dynamically allocate operation importance weights based on multiple factors, and set up a dynamic weight adjustment mechanism to regularly re-evaluate and adjust the weights; S4. Importance score calculation and verification: The score is calculated according to the behavior pattern and weight according to the relevant formula, and a verification mechanism is set up to compare with historical data to ensure that the score is accurate and reliable; S5. Collection granularity decision and optimization: The collection granularity adjustment strategy is determined based on the comparison between the score and the threshold, increasing or decreasing the collection information details, and the resource status of the terminal device is considered during the adjustment to ensure efficient and stable operation.

2. A method for collecting service operation information in a network information service system according to claim 1, characterized in that: The operation sequence buffer adopts an efficient data compression algorithm to reduce memory usage and improve reading and writing speeds, and automatically selects the optimal compression algorithm according to the data type and distribution characteristics of the operation information, such as Huffman coding compression for text and differential coding compression for numerical values.

3. The method for collecting service operation information in a network information service system according to claim 1, characterized in that: The machine learning classification algorithm adopts an integrated learning method, integrates the prediction results of multiple basic classification models, and determines the final classification result through voting or weighted average mechanism, thereby enhancing accuracy and robustness and reducing the risk of misjudgment.

4. A method for collecting service operation information in a network information service system according to claim 1, characterized in that: In the importance weight dynamic configuration step, the weight adjustment period is adaptively set according to the frequency of application business changes and the stability of user behavior. It is shortened when the business is changeable or the user behavior fluctuates greatly, and extended when the business is stable or the user behavior is fixed.

5. The method for collecting service operation information in a network information service system according to claim 1, characterized in that: The sequential pattern mining algorithm adopts parallel computing technology when calculating support, divides the operation sequence data into blocks to multiple computing cores or nodes for simultaneous calculation, and summarizes the results to obtain the final support to improve pattern recognition efficiency, which is suitable for large-scale data processing.

6. The method for collecting service operation information in a network information service system according to claim 1, characterized in that: It has corresponding specific adjustment strategies in different application scenarios: in the mobile news reading application scenario, the behavior pattern and score are determined according to the news browsing related operation type to adjust the news picture data collection; in the video live social platform scenario, the behavior pattern and score are determined according to the live interactive operation type to adjust the interactive information collection and transmission; in the enterprise-level collaborative office software scenario, the behavior pattern and score are determined according to the document collaboration operation type to adjust the document information collection and preprocessing.

7. A method for collecting service operation information in a network information service system according to claim 1, characterized in that: In the collection granularity adjustment decision step, when the collection granularity is adjusted, a corresponding collection strategy adjustment log is generated at the same time, recording the reason and time of the adjustment, detailed change information of the collection granularity before and after, and the estimated impact on device resources and network transmission, so as to provide reference for subsequent data analysis, troubleshooting and strategy optimization.

Images