A time-aware based service recommendation method

Abstract

The application relates to the fields of service recommendation and deep learning, in particular to a time-aware service recommendation method.A time-aware service recommendation method comprises the following steps: S1, data preprocessing; S2, calculating the similarity between users / services; S3, screening out a set of the first k users / services with the highest similarity to a target user / service; S4, predicting the QoS value of the target user; and S5, recommending services to the target user according to the final QoS prediction value.The application provides a time-aware service recommendation method for efficient and accurate service recommendation, which combines collaborative filtering and a long short-term memory network, fills part of the missing indicators by using the service quality indicators of adjacent time periods to alleviate the problem of data sparsity, screens out similar groups by using collaborative filtering, and learns the change trend of the service quality indicators related to similar users or similar services with the long short-term memory network over time.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of service recommendation and deep learning, and in particular, to a time-aware service recommendation method. BACKGROUND

[0002] Under the background of the continuous expansion of the Internet industry, a large amount of data is generated in daily life and production. The redundancy and complexity of a large amount of data bring many difficulties to both users and service providers. Users have difficulty in quickly and accurately finding services that meet their interests and needs, and managers have difficulty in reasonably allocating resources and accurately pushing to target users. The recommendation system effectively alleviates the above problems by closely linking users and services through specific ways, thereby bringing convenience to both service consumers and service providers.

[0003] One of the most important steps in service recommendation research is to accurately predict and calculate the lost or future QoS indicators through the current QoS indicators. The QoS indicators include multiple aspects, such as security, availability, accuracy, etc., and the QoS indicators are observed and provided by the network service program by combining services and service providers together. The service recommendation system can recommend services to users by predicting QoS indicators, and also can let service providers find user groups that will use their services and future service invocation states, therefore, improving the accuracy of QoS indicator prediction helps to analyze the interests and hobbies of users for services, thereby improving the performance of service recommendation.

[0004] The most important step in the recommendation system is the recommendation algorithm used. With the continuous updating and improvement of the recommendation system, more and more recommendation algorithms are proposed and applied in different fields. One of the most popular and easily extended recommendation algorithms in the service recommendation system is the collaborative filtering (CF) therein. CF can mine the hidden preferences of users and use similar users or similar items for recommendation.

[0005] However, CF has difficulty in dealing with data sparsity problems, and CF also has difficulty in mining the changes of user behavior in time series. This means that in the Internet environment where the number of users and services is rapidly increasing, CF often does not achieve good results in service recommendation. At the same time, since the behavior of users will change due to the change of time, CF also has difficulty in learning the impact of this change on service recommendation. SUMMARY

[0006] The purpose of the present application is to provide a time-aware service recommendation method based on collaborative filtering and long short-term memory network, which fills in part of the missing indicators by using the quality of service indicators of adjacent time periods to alleviate the problem of data sparsity, filters similar groups by using collaborative filtering, and learns the trend of the quality of service indicators related to similar users or similar services over time by using long short-term memory network, so as to realize efficient and accurate service recommendation.

[0007] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0008] A time-aware service recommendation method based on time-aware service recommendation method, comprising the following steps:

[0009] S1, data preprocessing, obtaining user information from a cloud server to form a data packet, and preprocessing the data packet;

[0010] S2, calculating the similarity between users / services according to the QoS value generated by the user calling service according to the data packet preprocessed in step S1;

[0011] S3, combining the similarity of users / services obtained in step S2 with the geographical location information corresponding to the user / service, and screening out the top k user / service set with the highest similarity to the target user / service;

[0012] S4, combining the similar target user / service set obtained in step S3, training the long short-term memory network model of the QoS value of the target user, and predicting the QoS value of the target user;

[0013] S5, combining the prediction result of step S4, calculating the final QoS prediction value, and recommending services to the target user according to the final QoS prediction value.

[0014] Further, the present application further comprises the following technical solutions:

[0015] The data preprocessing of step S1 comprises the following steps:

[0016] S101, data normalization, using linear normalization method to convert the range of QoS value to , the specific formula is as follows:

[0017]

[0018] Wherein, represents the user u calling the service t at i the QoS indicator generated by , Max and Min represent the maximum and minimum values of all QoS indicators, respectively;

[0019] S102, fill in the missing values with the QoS values in the adjacent time period of the missing values to alleviate the problem of data sparseness, and the specific calculation formula is:

[0020]

[0021] Among them, represents the average value of the QoS indicators of the corresponding service called by the user in the T time period, which is used to fill in the with a value of 0 or null.

[0022] Among them, the definition of T is:

[0023]

[0024] Among them, 0 and 63 represent the value range of the time slice, and d represents the length of the adjacent time period before and after.

[0025] The similarity calculation between the user / service in the step S2 includes the following steps:

[0026] S201, propose a personalized weight based on the number of services called by the user, the more services called by the user, the greater the weight of the user, and the specific calculation formula is:

[0027]

[0028] Among them, represents the total number of all services, represents the number of services called by the user, represents the influence degree of this weight in the whole model, when 0, it can be considered that the weight does not affect the model;

[0029] S202, calculate the importance weight, the more the number of users who call two services, the more important the service to the target service, and the specific calculation formula is:

[0030]

[0031] Among them, represents the number of users who call service , represents the number of users who call service and service .

[0032] S203, calculate the interest decay coefficient of the service by using Newton cooling theorem, and the specific calculation formula is:

[0033]

[0034] wherein, represents the time when the user last invoked the service represents the degree of interest of the user in the service at the time represents the decay coefficient;

[0035] S204, combined with the above three weights and coefficients, the Pearson correlation coefficient is used to calculate the similarity between services, and the specific calculation formula is:

[0036]

[0037]

[0038] wherein, U is , represents the user set that has invoked the service at the time represents the average of the QoS indicators generated by the users in invoking the service in the time

[0039] The step S203 further comprises: if the initial degree of interest is the same each time the user invokes the service, the weight is simplified to .

[0040] The similarity calculation formula between users in the step S2 is:

[0041]

[0042] wherein, indicates the number of all users, indicates the number of users who have invoked the service, represents the influence degree of the weight in the whole model;

[0043]

[0044] wherein, represents the number of services invoked by the user represents the services invoked by the user and the user

[0045] ​​​​​​​​

[0046] wherein, represents the time when the service was last invoked by the user, represents the time when the service was last invoked by the user, represents the time when the service was last invoked by the user, represents the degree of interest of the user for the service, represents the time when the service was last invoked by the user, represents the time when the service was last invoked by the user, represents the degree of interest of the user for the service, represents the decay coefficient, represents the natural constant;

[0047]

[0048]

[0049] wherein, S is represents the set of services that were invoked by the user together with at the time instant , at the time instant represents the average of the QoS indicators generated by the services in the set at the time instant .

[0050] The set of the top k users / services similar to the target user / service includes the following steps:

[0051] S301, according to the autonomous domain, the user / service in the same autonomous domain as the target user / service is screened, if the number of similar users / services is less than k , the screening range is expanded, and the next step is entered; if the number of similar users / services is greater than k , the similar users / services are screened in this set by using the Top-k algorithm, and the users / services with a similarity less than 0 are excluded;

[0052] if the number of users after screening is less than k , the screening range is expanded; if the number of users after screening is not less than k , the similar user / service set is returned;

[0053] S302, according to the country, the user / service in the same country as the target user / service is screened, if the number of similar users / services is less than k , the screening range is expanded, and the next step is entered; if the number of similar users / services is greater than k , the similar users / services are screened in this set by using the Top-k algorithm, and the users / services with a similarity less than 0 are excluded;

[0054] if the number of users after screening is less than kIf the number of users after the screening is not less than k If the number of users after the screening is not less than

[0055] S303, using all data, calculating the similarity with all other users / services, using Top-k algorithm to screen out the first k If the number of users after the screening is not less than k If the number of users after the screening is not less than

[0056] The model training of the long short-term memory network in the step S4 and the prediction of the QoS value include the following steps:

[0057] S401, constructing a related data set according to the similar users of the target user and the target user itself, and training a long short-term memory network model using the data set to obtain the QoS value generated by the target user when calling the target service at a specific time;

[0058] S402, constructing a related data set according to the similar services of the target service and the target service itself, and training another long short-term memory network model using the data set to obtain the QoS value generated by the target user when calling the target service at a specific time;

[0059] S403, performing reverse normalization processing on the predicted value, and the calculation formula is:

[0060]

[0061] Wherein, and respectively represent the maximum and minimum values of all QoS indicators.

[0062] The calculation process of the final QoS prediction value and the service recommendation in the step S5 include the following steps:

[0063] S501, calculating the average similarity weight of users using the similarity between users, and the calculation formula is:

[0064]

[0065] Wherein, represents the length of the adjacent time period when filling in the missing value;

[0066] S502, calculating the user-based prediction value using the QoS value predicted by the similar users, and the calculation formula is:

[0067]

[0068] Wherein, represents the user-based prediction value of the similar user obtained through the LSTM modelu exist Call the service at any time i The generated QoS prediction value, Similar users during LSTM training and prediction Call the service at any time i The generated QoS value;

[0069] S503 calculates the average service similarity weight using the similarity between users. The calculation formula is as follows:

[0070]

[0071] in, This represents the length of the adjacent time interval used when filling in missing values;

[0072] S504, calculates the service-based prediction value using the QoS values ​​predicted by similar services, using the following formula:

[0073]

[0074] in, Represents users based on similar services obtained through the LSTM model. u exist Call the service at any time i The generated QoS prediction value, Similar services during LSTM training and prediction Always by users u The QoS value generated by the call;

[0075] S505 uses a weighted approach, combining user-based and service-based prediction methods, to calculate the final QoS prediction value. The calculation formula is as follows:

[0076]

[0077] in, This represents the proportion of user-based LSTM prediction methods versus service-based LSTM prediction methods.

[0078] S506 selects the best-performing service based on the predicted QoS value and recommends it to the target user.

[0079] The beneficial effects of adopting the technical solution of this invention are as follows:

[0080] 1. This invention utilizes QoS values ​​from adjacent time periods to fill in some missing QoS values, effectively alleviating the problem caused by data sparsity and improving the accuracy of prediction.

[0081] 2. This invention uses Newton's law of cooling to simulate changes in user interest in services, thus better taking into account the actual situation of user service calls.

[0082] 3. By incorporating personalized and importance weights when calculating similarity, the impact of users or services with limited data on recommendations is effectively reduced.

[0083] 4. By using Long Short-Term Memory (LSTM) networks for QoS value prediction, the patterns of QoS changes over time are effectively learned, thus improving the accuracy of service recommendations. Detailed Implementation

[0084] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0085] This invention provides a time-aware service recommendation method that combines collaborative filtering and long short-term memory networks to improve the accuracy of service recommendations. The method specifically includes the following steps:

[0086] S1. Data preprocessing: The collected data is preprocessed to reduce errors introduced by the data itself and to alleviate the problem of data sparsity to some extent. This includes the following steps:

[0087] S101, Data Normalization: Using linear normalization, the range of QoS values ​​is transformed to... The specific formula is as follows:

[0088]

[0089] Among them, representing users u exist t Call the service at any time i The resulting QoS metrics and These represent the maximum and minimum values ​​of all QoS metrics, respectively.

[0090] S102. Fill missing values ​​with QoS values ​​within adjacent time periods to alleviate data sparsity. The specific calculation formula is as follows:

[0091]

[0092] in, This represents the average QoS metric for the corresponding service accessed by a user within the time period T, used to fill in values ​​of 0 or empty. Where T is defined as:

[0093]

[0094] Here, 0 and 63 refer to the range of values ​​for the time slice, and d represents the length of adjacent time periods. By adjusting d, the range of data on which missing values ​​depend can be adjusted, thereby adjusting the prediction performance of the entire model to adapt to data with different sparsity and application scenarios.

[0095] S2. Calculate the similarity between users (services) based on the QoS values ​​generated by user service calls, specifically including the following steps:

[0096] S201. Since different users contribute differently to service recommendations in real life, to adjust the contribution of different users to similarity, a personalized weight based on the number of services a user invokes is proposed. The more services a user invokes, the greater their weight. The specific calculation formula is as follows:

[0097]

[0098] in, This represents the total number of all services. This indicates the number of services a user has called. This represents the degree of influence of this weight in the entire model, when When the value is 0, it can be considered that the weight has no impact on the model. The example uses... The value is 0.9.

[0099] S202. To prevent services called by only a few users from affecting the selection of similar users, an importance weight is introduced. The more users who have called two services, the more important that service is to the target service. The specific calculation formula is as follows:

[0100]

[0101] in, This indicates that the service has been called. The number of users, This indicates that the service has been called. With services Users.

[0102] S203. Since user interests change over time, the time factor also affects the similarity calculation. This method uses Newton's law of cooling to calculate the service's interest decay coefficient. The specific calculation formula is as follows:

[0103]

[0104] in, represent The user's last call to the service At that moment, Representative at When users are using the service The degree of interest This represents the decay coefficient. Here, we assume that the initial level of interest is the same each time a user calls the service, so this weight can be simplified to... The example uses The value is 0.2.

[0105] S204. Combining the above three weights and coefficients, the Pearson correlation coefficient is used to calculate the similarity between services. The specific calculation formula is as follows:

[0106]

[0107]

[0108] in, U that is , representing in The service was called frequently and services The user set, Representative at All within a moment Internal user call service The average of the generated QoS metrics.

[0109] Furthermore, the specific steps for calculating similarity between users are as follows:

[0110] S205. Similar to the calculation method for service similarity, the specific formula for calculating user similarity is as follows:

[0111]

[0112]

[0113]

[0114]

[0115]

[0116] in, This represents the total number of users. This indicates the number of users who have used the service. This represents the degree of influence of this weight in the entire model. This represents the last time the service was accessed by a user. Representative at Time user The degree of interest in the service This represents the attenuation coefficient. S is... , representing in Time users and A collection of services that have been called together. On behalf of users Call within a time period The average QoS metrics generated by services within the set.

[0117] SS3. Based on similarity and geographic location information, filter out the top users (services) with the highest similarity to the target user (service). k A collection of users (services) specifically includes the following steps:

[0118] S301, filter users (services) that are in the same autonomous region as the target user (service) based on the autonomous region. If the number of similar users (services) is less than... k If the number of similar users (services) is greater than [a certain threshold], then expand the filtering scope and proceed to the next step; k Within this set, the Top-k algorithm is used to filter out similar users (services), excluding users (services) with a similarity score less than 0. If the number of users after filtering is less than [a certain threshold], [the algorithm will fail]. k If the number of users after filtering is not less than [a certain threshold], then expand the filtering scope; k If so, it returns a set of similar users (services).

[0119] S302. Filter users (services) in the same country as the target user (service) based on their country of origin. If the number of similar users (services) is less than [a certain number], [further details are needed]. k If the number of similar users (services) is greater than [a certain threshold], then expand the filtering scope and proceed to the next step; k Within this set, the Top-k algorithm is used to filter out similar users (services), excluding users (services) with a similarity score less than 0. If the number of users after filtering is less than [a certain threshold], [the algorithm will fail]. k If the number of users after filtering is not less than [a certain threshold], then expand the filtering scope; k If so, it returns a set of similar users (services).

[0120] S303. Using all data, calculate the similarity with all other users (services), and use the Top-k algorithm to filter out the top... k If there are users (services) with a similarity of not less than 0, and if they are not satisfied... k They only use these similar users (services).

[0121] S4. Train the Long Short-Term Memory network model using QoS values ​​associated with the target user and similar users, and predict the target QoS value. This includes the following steps:

[0122] S401. Construct a relevant dataset based on similar users of the target user and the target user itself, and use the dataset to train a Long Short-Term Memory (LSTM) network model to obtain the QoS value generated when the target user calls the target service at a specific time. The learning rate used in this example is 0.1, and the number of hidden neurons is 25.

[0123] S402. Construct a relevant dataset based on similar services to the target service and the target service itself. Use this dataset to train another Long Short-Term Memory (LSTM) network model to obtain the QoS value generated when a target user calls the target service at a specific time. The learning rate used in this example is 0.1, and the number of hidden neurons is 25.

[0124] S403. Perform inverse normalization on the predicted values. The specific calculation formula is as follows:

[0125]

[0126] in, and These represent the maximum and minimum values ​​of all QoS metrics, respectively.

[0127] S5. Combining the results from S4, calculate the final QoS prediction value and recommend services to the target user based on the results. This includes the following steps:

[0128] S501. Calculate the average user similarity weight using the similarity between users. The specific calculation formula is as follows:

[0129]

[0130] in, This represents the length of the adjacent time period used when filling in missing values.

[0131] S502. Calculate the user-based predicted value using the QoS values ​​predicted by similar users. The specific calculation formula is as follows:

[0132]

[0133] in, Represents users based on similar users obtained through the LSTM model. u exist Call the service at any time i The generated QoS prediction value, Similar users during LSTM training and prediction Call the service at any time i The generated QoS value.

[0134] S503. Calculate the average service similarity weight using the similarity between users. The specific calculation formula is as follows:

[0135]

[0136] in, This represents the length of the adjacent time period used when filling in missing values.

[0137] S504. Calculate the service-based prediction value using the QoS values ​​predicted by similar services. The specific calculation formula is as follows:

[0138]

[0139] in, Represents users based on similar services obtained through the LSTM model. u exist Call the service at any time i The generated QoS prediction value, Similar services during LSTM training and prediction Always by users u The QoS value generated by the call.

[0140] S505. Using a weighted approach, combining user-based prediction methods and service-based prediction methods, the final QoS prediction value is calculated. The specific calculation formula is as follows:

[0141]

[0142] in, This represents the proportion of user-based LSTM prediction methods versus service-based LSTM prediction methods.

[0143] S506. Based on the predicted QoS value, select the best-performing service (short response time and high throughput) and recommend it to the target user.

[0144] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A time-aware service recommendation method, characterized in that, Includes the following steps: S1. Data preprocessing: Obtain response time and throughput data from the cloud server through the service quality log information generated by user service calls. At the same time, obtain the IP address, autonomous system, city, and country information of the service and user. Integrate all data and information into a data packet and preprocess the data packet. S2. Calculate the similarity between users and services based on the QoS values ​​generated by the user's service calls after preprocessing the data packets in step S1. S3. Combine the user / service similarity obtained in step S2 with the geographical location information corresponding to the user / service to filter out the set of the top k users / services with the highest similarity to the target user / service; S4. Based on the set of similar target users / services obtained in step S3, train a long short-term memory network model for the QoS values ​​of the target users and predict the QoS values ​​of the target users. S5. Based on the prediction results of step S4, calculate the final QoS prediction value, and recommend services to the target user according to the final QoS prediction value. The similarity calculation between users / services in step S2 includes the following steps: S201. A personalized weighting based on the number of services a user invokes is proposed. The more services a user invokes, the greater their weight. The specific calculation formula is as follows: ， in, This represents the total number of all services. This indicates the number of services a user has called. This represents the degree of influence of this weight in the entire model, when When the value is 0, it can be considered that the weight has no impact on the model; S202. Calculate the importance weight. The more users who have called both services, the more important the service is to the target service. The specific calculation formula is as follows: ， in, This indicates that the service has been called. The number of users, This indicates that the service has been called. With services Users; S203. Calculate the interest decay coefficient of the service using Newton's law of cooling. The specific calculation formula is as follows: ， in, represent The user's last call to the service At that moment, Representative at When users are using the service The degree of interest Represents the attenuation coefficient; S204. Combining the above three weights and coefficients, the Pearson correlation coefficient is used to calculate the similarity between services. The specific calculation formula is as follows: ， ， in, U that is , representing in The service was called frequently and services The user set, Representative at All within a moment Internal user call service The average value of the generated QoS metrics; Step S203 further includes: if the initial level of interest is the same each time a user calls the service, simplify the weights to... ; The formula for calculating the similarity between users in step S2 is as follows: ， in, This represents the total number of users. This indicates the number of users who have used the service. This represents the degree of influence of this weight in the entire model; ， in, On behalf of users The number of services called. Representing the user With users Services that have been called; ， in, This represents the last time the service was accessed by a user. Representative at Time user The degree of interest in the service Representative at Time user The degree of interest in the service Represents the attenuation coefficient. Represents the natural constant; ， ， Where S is , representing in Time users and A collection of services that have been called together. On behalf of users Call within a time period The average QoS metrics generated by services within the set.

2. The time-aware service recommendation method according to claim 1, characterized in that, The data preprocessing in step S1 includes the following steps: S101. Data normalization: Using linear normalization, the range of QoS values ​​is transformed to... The specific formula is as follows: ， in, On behalf of users u exist t Call the service at any time i The resulting QoS metrics and These represent the maximum and minimum values ​​of all QoS metrics, respectively. S102. Fill missing values ​​with QoS values ​​within adjacent time periods to alleviate data sparsity. The specific calculation formula is as follows: ， in, This represents the average QoS metric for the corresponding service accessed by a user within the time period T, used to fill in values ​​of 0 or empty. ; Where T is defined as: ， Here, 0 and 63 refer to the range of values ​​for the time slice, and d represents the length of adjacent time periods.

3. The time-aware service recommendation method according to claim 1, characterized in that, The process of selecting the set of the top k users / services with the highest similarity to the target user / service includes the following steps: S301. Filter out users / services in the same autonomous region as the target user / service based on the autonomous region. If the number of similar users / services is less than... k If the number of similar users / services is greater than [a certain threshold], then expand the filtering scope and proceed to the next step; k Within this set, the Top-k algorithm is used to filter out similar users / services and exclude users / services with a similarity of less than 0. If the number of users after filtering is less than k If the number of users after filtering is not less than [a certain threshold], then expand the filtering scope; k If so, it returns a set of similar users / services; S302. Filter users / services in the same country as the target user / service based on their country of origin. If the number of similar users / services is less than [a certain threshold], [further details are needed]. k If the number of similar users / services is greater than [a certain threshold], then expand the filtering scope and proceed to the next step; k Within this set, the Top-k algorithm is used to filter out similar users / services and exclude users / services with a similarity of less than 0. If the number of users after filtering is less than k If the number of users after filtering is not less than [a certain threshold], then expand the filtering scope; k If so, it returns a set of similar users / services; S303. Using all data, calculate the similarity to all other users / services, and use the Top-k algorithm to filter out the top... k A number of users / services with a similarity of at least 0, if not satisfied k They only use these similar users / services.

4. The time-aware service recommendation method according to claim 1, characterized in that, Step S4, which involves training the Long Short-Term Memory network model and predicting QoS values, includes the following steps: S401. Construct a relevant dataset based on similar users of the target user and the target user itself, and use the dataset to train a long short-term memory network model to obtain the QoS value generated when the target user calls the target service at a specific time. S402. Construct a relevant dataset based on similar services of the target service and the target service itself, and use the dataset to train another long short-term memory network model to obtain the QoS value generated when the target user calls the target service at a specific time. S403. Perform inverse normalization on the predicted values. The calculation formula is as follows: ， in, and These represent the maximum and minimum values ​​of all QoS metrics, respectively.

5. The time-aware service recommendation method according to claim 1, characterized in that, The calculation process for the final QoS prediction value and service recommendation in step S5 includes the following steps: S501. Calculate the average user similarity weight using the similarity between users. The calculation formula is as follows: ， in, This represents the length of the adjacent time interval used when filling in missing values; S502. Calculate the user-based predicted value using the QoS values ​​predicted by similar users. The calculation formula is as follows: ， in, Represents users based on similar users obtained through the LSTM model. u exist Call the service at any time i The generated QoS prediction value, Similar users during LSTM training and prediction Call the service at any time i The generated QoS value; S503 calculates the average service similarity weight using the similarity between users. The calculation formula is as follows: ， in, This represents the length of the adjacent time interval used when filling in missing values; S504, calculates the service-based prediction value using the QoS values ​​predicted by similar services, using the following formula: ， in, Represents users based on similar services obtained through the LSTM model. u exist Call the service at any time i The generated QoS prediction value, Similar services during LSTM training and prediction Always by users u The QoS value generated by the call; S505 uses a weighted approach, combining user-based and service-based prediction methods, to calculate the final QoS prediction value. The calculation formula is as follows: ， in, This represents the proportion of user-based LSTM prediction methods versus service-based LSTM prediction methods. S506 selects the best-performing service based on the predicted QoS value and recommends it to the target user.

Images