Service recommendation quality evaluation method and system

Abstract

The invention discloses a service recommendation quality evaluation method and system, and relates to the technical field of service recommendation systems, and the method comprises the steps: collecting and processing multi-dimensional data of users and service resources in a recommendation scene, and outputting a time-serialized multi-dimensional data set; based on the multi-dimensional data set, defining causal relationship variables of service recommendation quality evaluation, and determining a causal relationship path among the variables by using a causal inference theory; constructing a user behavior causal inference model by combining a causal inference theory and a real-time feedback mechanism; based on a user behavior causal inference model, performing personalized grouping on the users, and calculating causal weights of the users for different recommendation factors; a recommendation strategy is dynamically adjusted based on personalized grouping and causal weights of the users; according to the method, the user behavior causal inference model is constructed by combining logistic regression and L2 regularization, and the prediction stability and generalization ability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of service recommendation quality evaluation, and in particular to a service recommendation quality evaluation method and system. Background Art

[0002] As an important part of information retrieval and personalized services, service recommendation systems have been widely studied and applied in recent years. With the popularization of the Internet and mobile devices, users' demand for personalized recommendations is growing, prompting the continuous evolution of recommendation technology.

[0003] Although existing recommendation systems have made significant technological progress, they still have some shortcomings in practical applications. In particular, when dealing with complex and changeable user behaviors and situations, existing technologies often have low accuracy and are difficult to meet personalized needs. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a service recommendation quality evaluation method to solve the accuracy and personalization problems of service recommendations in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a service recommendation quality evaluation method, which comprises: Collect and process the multidimensional data of users and service resources in the recommendation scenario, and output a time-series multidimensional data set; Based on the multidimensional data set, the causal relationship variables of service recommendation quality evaluation are defined, and the causal relationship paths between the variables are clarified by using the causal inference theory; According to the causal relationship path, causal effect analysis is performed on the variables on the causal path, and preliminary causal effect values ​​are output; By combining causal inference theory and real-time feedback mechanism, a user behavior causal inference model is constructed; Based on the user behavior causal inference model, users are grouped individually and the causal weights of users for different recommendation factors are calculated; Dynamically adjust the recommendation strategy based on users' personalized grouping and causal weights.

[0007] As a preferred solution of the service recommendation quality evaluation method of the present invention, wherein: the multidimensional data of users and service resources in the recommendation scenario is collected and processed, and a time-series multidimensional data set is output. The specific steps are as follows: Use a unified data access interface to collect user behavior data, service feature data, and user context data in real time; Use distributed processing framework to process data; Based on rule filtering and logic verification, identify illogical and abnormal noise data in user behavior data, service feature data, and user context data; Through interpolation, the missing behaviors in the user behavior data time series are filled using the mean of behaviors at previous and subsequent time points. The missing values ​​in the service feature data are calculated using the collaborative feature filling method based on the mean of similar features of other services. The missing geographic location data in the user context data is filled using the user's last known location. Use the Z-score standardization method to normalize user behavior data, service feature data, and user context data. Convert multidimensional data into a unified standard format; Organize user behavior data into time series according to the chronological order of user behavior; By associating the service ID in the user behavior record with the primary key of the service feature data, the service feature data corresponding to the behavior data is supplemented; According to the timestamp of user behavior, match the user context data in the corresponding time period and supplement the context information; Each record contains user behavior data, service feature data, and user context data, and is sorted by time to obtain a time-serialized multidimensional dataset.

[0008] As a preferred solution of the service recommendation quality evaluation method of the present invention, wherein: based on the multidimensional data set, the causal relationship variables of the service recommendation quality evaluation are defined, and the causal relationship path between the variables is clarified by using the causal inference theory. The specific steps are as follows: Based on the multidimensional data set, the final result of user behavior is defined as the dependent variable Y, including whether to purchase, the user's rating result after purchase, and the user's total stay time; Take the recommended content variable X1, the user context variable X2, and the user history behavior variable X3 as the independent variable X; Treat each variable as a node in a causal graph; Through causal inference theory and data correlation analysis, the causal relationship path between variables is determined, and the recommended content variable X1 directly affects the final result Y of user behavior; User context variables X2 and user historical behavior variables X3 indirectly affect the final result Y of user behavior by adjusting preferences for recommended content; Construct a cause-effect diagram to clarify the cause-effect path between variables.

[0009] As a preferred solution of the service recommendation quality evaluation method of the present invention, wherein: according to the causal relationship path, the variables on the causal path are analyzed for causal effects, and a preliminary causal effect value is output. The specific steps are as follows: According to the causal relationship path in the causal diagram, the independent variables are divided into direct causal variables that directly affect the dependent variable Y, indirect causal variables that indirectly affect the dependent variable Y through direct causal variables, and irrelevant variables that do not appear in the causal relationship path; Keep direct and indirect causal variables and eliminate irrelevant variables; According to the values ​​of the key variables in the independent variables, the recommended content variable X1, the user context variable X2 and the user historical behavior variable X3 are grouped, with obvious differences between the groups, while maintaining the consistency of other independent variables; The group with high values ​​of each key variable in the independent variable is used as the experimental group, and the group with low values ​​of the key variable is used as the control group; Calculate the mean difference of the dependent variable Y between the experimental group and the control group; The group difference value of each key variable is taken as the preliminary causal effect value of the corresponding key causal variable.

[0010] As a preferred solution of the service recommendation quality evaluation method of the present invention, wherein: the user behavior causal inference model is constructed by combining causal inference theory and real-time feedback mechanism, and the specific steps are as follows: Based on the logistic regression model in causal inference theory, the probability of users receiving recommended content is calculated, and a user behavior causal inference model is constructed. The expression is as follows: ; in, is the probability of the user receiving the recommended content variable X1, is the intercept term, is the coefficient of the user context variable X2, is the coefficient of the user's historical behavior variable X3; Mix the user context variable X2, the user history behavior variable X3, and the label of whether the user accepts the recommended content variable X1, and divide them into a training set and a validation set; Input the training set into the user behavior causal inference model for training; The log-likelihood function with L2 regularization term introduced by maximizing , the expression is as follows: ; in, is the total number of users, is the index of the number of users, For the The probability that a user accepts the recommendation, For the A label of whether to accept the recommended content variable X1, is the regularization coefficient, ; During the iterative training of the user causal inference model, the gradient direction is updated , get the best ; When the maximum number of iterations is reached, the training is terminated and the validation set is input to verify the performance of the user behavior causal inference model by calculating the accuracy, recall rate, and F1 score of the user behavior causal inference model. Set the probability threshold, when When it is greater than the probability threshold, the user is considered to have accepted the recommendation, otherwise, the user is considered to have not accepted the recommendation; Based on the probability calculation results of users receiving recommended content, users who accepted the recommendation are used as the experimental group, and users who did not accept the recommendation are used as the control group; The nearest neighbor matching method is used to match users according to the Euclidean distance of the probability of users receiving recommended content, and the experimental group users are matched with the control group users whose probability of users receiving recommended content is closest to that of users in the experimental group. After the matching is completed, the mean difference of the dependent variable Y in the experimental group and the control group is calculated as the causal effect value of the recommended content on the user's decision; Collect the latest user behavior data in real time, recalculate the probability of the user receiving the recommended content variable X1, and dynamically adjust the probability threshold based on the latest probability of the user receiving the recommended content.

[0011] As a preferred solution of the service recommendation quality evaluation method of the present invention, the specific steps are as follows: based on the user behavior causal inference model, users are personalized grouped, and the causal weights of users for different recommendation factors are calculated. The recommended content variable X1, user context variable X2 and user historical behavior variable X3 are used as key features for clustering; By plotting the change curve of the sum of squared clustering errors (SSE) with the number of clusters k, the k value at the inflection point is selected as the optimal k value; Use K-Means algorithm to cluster users and divide them into k different clusters; Input the recommended content variable X1, user context variable X2 and user historical behavior variable X3 of each cluster into the user behavior causal inference model to obtain the probability of users in each cluster receiving the recommended content; The boundary effect values ​​in the user context variable X2 and the user historical behavior variable X3 are calculated as personalized weights to express the influence of user behavior in the cluster. The expression is as follows: ; in, is the personalized weight of the user context variable X2 and the user historical behavior variable X3, ; Normalize the personalization weights.

[0012] As a preferred solution of the service recommendation quality evaluation method of the present invention, wherein: the recommendation strategy is dynamically adjusted based on the personalized grouping and causal weight of the user, and the specific steps are as follows: Collect user context data and user behavior data in real time through a unified data access interface; Update the personalized weights of the user context variable X2 and the user historical behavior variable X3 based on the real-time user context data and user behavior data; For each service item, the product of the updated personalized weight and the user context variable X2 and the user historical behavior variable X3 is calculated as the comprehensive score of the service item; Optimize recommendation strategies according to different types of users.

[0013] In a second aspect, the present invention provides a service recommendation quality evaluation system, comprising, including, a data acquisition module, a causal relationship path module, a causal effect analysis module, a model building module, a causal weight calculation module and a recommendation strategy adjustment module; The data collection module is used to collect and process the multidimensional data of users and service resources in the recommendation scenario and output a time-series multidimensional data set; The causal relationship path module is used to define the causal relationship variables of the service recommendation quality evaluation based on the multidimensional data set, and to clarify the causal relationship path between the variables by using the causal inference theory; The causal effect analysis module is used to perform causal effect analysis on the variables on the causal path according to the causal relationship path, and output preliminary causal effects; The model building module is used to build a user behavior causal inference model by combining causal inference theory and real-time feedback mechanism; The causal weight calculation module is used to perform personalized grouping of users based on the user behavior causal inference model and calculate the causal weights of users for different recommendation factors; The recommendation strategy adjustment module is used to dynamically adjust the recommendation strategy based on the personalized grouping and causal weight of the user.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the service recommendation quality evaluation method described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the service recommendation quality evaluation method as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are as follows: by collecting and processing multidimensional data of user behavior, service characteristics and user context, and outputting time-series multidimensional data sets, the real-time and accuracy of the recommendation system are ensured; based on causal inference theory, the causal relationship path between variables is clarified, and the influence of different factors on user behavior is quantified through causal effect analysis, thereby avoiding false correlations and deviations; the constructed user behavior causal inference model combines logistic regression and L2 regularization, which improves the stability and generalization ability of the prediction; by personalized grouping of users and calculating causal weights, the system can provide more accurate recommendation services for different types of users, dynamically adjust the recommendation strategy to meet the immediate needs of users, and the real-time feedback mechanism and incremental learning algorithm enable the system to quickly respond to changes in user behavior and continuously optimize the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 This is a flow chart of the service recommendation quality evaluation method in Example 1.

[0019] Figure 2 This is a module diagram of the service recommendation quality evaluation system in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a service recommendation quality evaluation method, comprising the following steps: Collect and process the multidimensional data of users and service resources in the recommendation scenario, and output a time-series multidimensional data set; Based on the multidimensional data set, the causal relationship variables of service recommendation quality evaluation are defined, and the causal relationship paths between the variables are clarified by using the causal inference theory; According to the causal relationship path, the variables on the causal path are analyzed for causal effects, and preliminary causal effects are output; By combining causal inference theory and real-time feedback mechanism, a user behavior causal inference model is constructed; Based on the user behavior causal inference model, users are grouped individually and the causal weights of users for different recommendation factors are calculated; Dynamically adjust the recommendation strategy based on users' personalized grouping and causal weights.

[0024] S1. Collect and process the multidimensional data of users and service resources in the recommendation scenario, and output a time-series multidimensional data set.

[0025] Use a unified data access interface to collect user behavior data, service feature data, and user context data in real time; Specifically, through tracking technology and logs, user behavior data is recorded in real time, including clicks, stay time, purchase records, and ratings; The specific implementation of tracking can be based on an event-driven front-end framework (such as JavaScript monitoring user click events), and the data is transmitted to the back-end server for storage through the API interface; Extract service feature data from the platform service resource database, including service type, price, rating, and provider information; Collect user context data based on sensor data or environmental data from the user's device, including time, geographic location, device type, and network status; Use distributed processing frameworks (such as Apache Spark) to process data; Based on rule filtering and logic verification, identify illogical and abnormal noise data in user behavior data, service feature data, and user context data, for example, data with too short a retention time in user behavior data (such as click behavior of less than 1 second) to avoid false behavior from interfering with the model, and obviously wrong feature values ​​in service feature data (such as negative price), etc.; Through interpolation, the missing behaviors in the user behavior data time series are filled using the mean of behaviors at previous and subsequent time points. The missing values ​​in the service feature data are calculated using the collaborative feature filling method based on the mean of similar features of other services. The missing geographic location data in the user context data is filled using the user's last known location. Use the Z-score standardization method to normalize user behavior data, service feature data, and user context data. Convert multidimensional data into a unified standard format; Specifically, the timestamps of multidimensional data are uniformly converted to the ISO 8601 format standard, and the geographic locations are uniformly converted to longitude and latitude with six decimal places; Organize user behavior data into time series according to the chronological order of user behavior; By associating the service ID in the user behavior record with the primary key of the service feature data, the service feature data corresponding to the behavior data is supplemented; According to the timestamp of user behavior, match the user context data in the corresponding time period and supplement the context information; Each record contains user behavior data, service feature data, and user context data, and is sorted by time to obtain a time-serialized multidimensional dataset.

[0026] S2. Based on the multidimensional data set, define the causal relationship variables of service recommendation quality evaluation, and use causal inference theory to clarify the causal relationship path between the variables.

[0027] Based on the multidimensional data set, the final result of user behavior is defined as the dependent variable Y, including whether to purchase, the user's rating result after purchase, and the user's total stay time; Take the recommended content variable X1, the user context variable X2, and the user history behavior variable X3 as the independent variable X; Specifically, the recommended content variables include the recommended service type and service features (such as price, rating, provider information, etc.), the user context variables include context information such as time, geographic location, device type, network status, etc., and the user history behavior variables include the user's historical clicks, purchase records, ratings, and other behavioral features; Treat each variable as a node in a causal graph; Through causal inference theory and data correlation analysis, the causal relationship path between variables is determined, and the recommended content variable X1 directly affects the final result Y of user behavior; User context variables X2 and user historical behavior variables X3 indirectly affect the final result Y of user behavior by adjusting preferences for recommended content; Construct a cause-effect diagram to clarify the cause-effect path between variables.

[0028] S3. According to the causal relationship path, causal effect analysis is performed on the variables on the causal path, and preliminary causal effect values ​​are output.

[0029] According to the causal relationship path in the causal diagram, the independent variables are divided into direct causal variables that directly affect the dependent variable Y, indirect causal variables that indirectly affect the dependent variable Y through direct causal variables, and irrelevant variables that do not appear in the causal relationship path; Keep direct and indirect causal variables and eliminate irrelevant variables; According to the values ​​of key variables in the independent variables (such as service price, rating, user purchase, click, purchase time, location and frequency, etc.), the recommended content variable X1, user context variable X2 and user historical behavior variable X3 are grouped, with obvious differences between the groups, while maintaining the consistency of other independent variables; For example, the recommended content variables (such as service price) are grouped into low-price group (price lower than the median) and high-price group (price higher than the median), the user context variables (such as time period) are grouped into working hours (9:00-18:00) and non-working hours (18:00-24:00), and the user history behavior variables (such as purchase frequency) are grouped into high-frequency users (purchase frequency higher than the mean) and low-frequency users (purchase frequency lower than the mean). Other key variables such as service type and rating are basically the same; The group with high values ​​of each key variable in the independent variable is used as the experimental group, and the group with low values ​​of the key variable is used as the control group; Calculate the mean difference of the dependent variable Y between the experimental group and the control group; The group difference value of each key variable is taken as the preliminary causal effect value of the corresponding key causal variable.

[0030] S4. Construct a user behavior causal inference model by combining causal inference theory and real-time feedback mechanism.

[0031] Based on the logistic regression model in causal inference theory, the probability of users receiving recommended content is calculated, and a user behavior causal inference model is constructed. The expression is as follows: ; in, is the probability of the user receiving the recommended content variable X1, is the intercept term, which indicates the probability of the user accepting the recommendation when both X2 and X3 are 0. is the coefficient of the user context variable X2, which measures the influence of the user context variable on the user's acceptance of recommendations. is the coefficient of the user's historical behavior variable X3, which measures the influence of the user's historical behavior variable on the user's acceptance of recommendations; Mix the user context variable X2, the user history behavior variable X3, and the label of the user's acceptance of the recommended content variable X1 (binary result: 1 for acceptance, 0 for rejection), and divide them into a training set and a validation set; Furthermore, the training set is input into the user behavior causal inference model for training; The log-likelihood function with L2 regularization term introduced by maximizing , the expression is as follows: ; in, is the total number of users, is the index of the number of users, For the The probability that a user accepts the recommendation, For the A label of whether to accept the recommended content variable X1, is the regularization coefficient, ; During the iterative training of the user causal inference model, the gradient direction is updated , get the best ; When the maximum number of iterations is reached, the training is terminated and the validation set is input to verify the performance of the user behavior causal inference model by calculating the accuracy, recall rate, and F1 score of the user behavior causal inference model. Set the probability threshold, when When it is greater than the probability threshold, the user is considered to have accepted the recommendation, otherwise, the user is considered to have not accepted the recommendation; Specifically, design multiple experimental groups, each using a different probability threshold, compare the recommendation effects under different thresholds (such as user purchase rate, click-through rate, satisfaction, etc.), and select the threshold with the best effect as the final judgment standard; Based on the probability calculation results of users receiving recommended content, users who accepted the recommendation are used as the experimental group, and users who did not accept the recommendation are used as the control group; The nearest neighbor matching method is used to match users according to the Euclidean distance of the probability of users receiving recommended content, and the experimental group users are matched with the control group users whose probability of users receiving recommended content is closest to that of users in the experimental group. By matching, the difference between the experimental group and the control group is reduced, and the power of statistical tests is improved. This means that with the same sample size, the matched data can more accurately detect the real impact of recommended content on user decisions. After the matching is completed, the mean difference of the dependent variable Y in the experimental group and the control group is calculated as the causal effect value of the recommended content on the user's decision; Combined with the preliminary causal effect value, accurately estimate the actual impact of the recommended content on user decision-making, and continuously optimize the recommendation strategy based on the matching results, preliminary causal effect value, and changes in the causal effect value. For example, if the impact of a certain user context variable (such as price) on the recommendation effect gradually increases, this variable can be given more attention in future recommendations. Furthermore, the latest user behavior data is collected in real time, the probability of the user receiving the recommended content variable X1 is recalculated, and the probability threshold is dynamically adjusted according to the latest probability of the user receiving the recommended content.

[0032] S5. Based on the user behavior causal inference model, users are grouped individually and the causal weights of users for different recommendation factors are calculated.

[0033] The recommended content variable X1, user context variable X2 and user historical behavior variable X3 are used as key features for clustering; By plotting the change curve of the sum of squared clustering errors (SSE) with the number of clusters k, the k value at the inflection point is selected as the optimal k value; Use K-Means algorithm to cluster users and divide them into k different clusters; Specifically, a user is randomly selected as the first cluster center and initialized using the K-Means++ algorithm to ensure that the distance between the initial cluster centers is as large as possible; For each unselected user, calculate its minimum distance to the center of the selected cluster; Using the square of the minimum distance as the probability distribution, randomly select the next cluster center and repeat the above steps until k cluster centers are selected; Input the recommended content variable X1, user context variable X2 and user historical behavior variable X3 of each cluster into the user behavior causal inference model to obtain the probability of users in each cluster receiving the recommended content; The boundary effect values ​​in the user context variable X2 and the user historical behavior variable X3 are calculated as personalized weights to express the influence of user behavior in the cluster. The expression is as follows: ; in, is the personalized weight of the user context variable X2 and the user historical behavior variable X3, ; The personalized weights are normalized so that the sum of all weights is 1 to ensure the comparability of causal weights.

[0034] S6. Dynamically adjust the recommendation strategy based on the user's personalized grouping and causal weight.

[0035] Collect user context data and user behavior data in real time through a unified data access interface; Update the personalized weights of the user context variable X2 and the user historical behavior variable X3 based on the real-time user context data and user behavior data; For each service item, the product of the updated personalized weight and the user context variable X2 and the user historical behavior variable X3 is calculated as the comprehensive score of the service item for sorting and recommendation; Optimize recommendation strategies according to different types of users; Specifically, for example, for price-sensitive users (i.e., the causal weight of price is relatively large), cost-effective services are recommended first. The specific approach is to appropriately increase the weight of the price factor in the calculation of the comprehensive score so that services with lower prices and higher service quality are ranked higher.

[0036] This embodiment also provides a service recommendation quality evaluation system, including: a data collection module, a causal relationship path module, a causal effect analysis module, a model construction module, a causal weight calculation module and a recommendation strategy adjustment module; the data collection module is used to collect and process the multidimensional data of users and service resources in the recommendation scenario, and output a time-series multidimensional data set; the causal relationship path module is used to define the causal relationship variables for service recommendation quality evaluation based on the multidimensional data set, and use the causal inference theory to clarify the causal relationship path between each variable; the causal effect analysis module is used to perform causal effect analysis on the variables on the causal path according to the causal relationship path, and output preliminary causal effects; the model construction module is used to construct a user behavior causal inference model by combining the causal inference theory and the real-time feedback mechanism; the causal weight calculation module is used to perform personalized grouping of users based on the user behavior causal inference model, and calculate the causal weights of users for different recommendation factors; the recommendation strategy adjustment module is used to dynamically adjust the recommendation strategy based on the personalized grouping and causal weights of users.

[0037] This embodiment also provides a computer device, which is suitable for the service recommendation quality evaluation method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the service recommendation quality evaluation method proposed in the above embodiment.

[0038] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0039] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the service recommendation quality evaluation method proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0040] In summary, the present invention collects and processes multidimensional data of user behavior, service characteristics and user context, outputs time-series multidimensional data sets, ensures the real-time and accuracy of the recommendation system, clarifies the causal relationship path between variables based on causal inference theory, quantifies the impact of different factors on user behavior through causal effect analysis, avoids false correlations and deviations, and constructs a user behavior causal inference model that combines logistic regression and L2 regularization to improve the stability and generalization ability of the prediction. By grouping users individually and calculating causal weights, the system can provide more accurate recommendation services for different types of users, dynamically adjust recommendation strategies to meet users' immediate needs, and the real-time feedback mechanism and incremental learning algorithm enable the system to quickly respond to changes in user behavior and continuously optimize the recommendation effect.

[0041] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A service recommendation quality evaluation method, characterized in that: include, Collect and process the multidimensional data of users and service resources in the recommendation scenario, and output a time-series multidimensional data set; Based on the multidimensional data set, the causal relationship variables of service recommendation quality evaluation are defined, and the causal relationship paths between the variables are clarified by using the causal inference theory; According to the causal relationship path, causal effect analysis is performed on the variables on the causal path, and preliminary causal effect values ​​are output; By combining causal inference theory and real-time feedback mechanism, a user behavior causal inference model is constructed; Based on the user behavior causal inference model, users are grouped individually and the causal weights of users for different recommendation factors are calculated; Dynamically adjust the recommendation strategy based on users' personalized grouping and causal weights.

2. The service recommendation quality evaluation method according to claim 1, characterized in that: The multidimensional data of users and service resources in the recommendation scenario are collected and processed to output a time-series multidimensional data set. The specific steps are as follows: Use a unified data access interface to collect user behavior data, service feature data, and user context data in real time; Use distributed processing framework to process data; Based on rule filtering and logic verification, identify illogical and abnormal noise data in user behavior data, service feature data, and user context data; Through interpolation, the missing behaviors in the user behavior data time series are filled using the mean of behaviors at previous and subsequent time points. The missing values ​​in the service feature data are calculated using the collaborative feature filling method based on the mean of similar features of other services. The missing geographic location data in the user context data is filled using the user's last known location. Use the Z-score standardization method to normalize user behavior data, service feature data, and user context data. Convert multidimensional data into a unified standard format; Organize user behavior data into time series according to the chronological order of user behavior; By associating the service ID in the user behavior record with the primary key of the service feature data, the service feature data corresponding to the behavior data is supplemented; According to the timestamp of user behavior, match the user context data in the corresponding time period and supplement the context information; Each record contains user behavior data, service feature data, and user context data, and is sorted by time to obtain a time-serialized multidimensional dataset.

3. The service recommendation quality evaluation method according to claim 2, characterized in that: Based on the multidimensional data set, the causal relationship variables of service recommendation quality evaluation are defined, and the causal relationship path between the variables is clarified by using causal inference theory. The specific steps are as follows: Based on the multidimensional data set, the final result of user behavior is defined as the dependent variable Y, including whether to purchase, the user's rating result after purchase, and the user's total stay time; Take the recommended content variable X1, the user context variable X2, and the user history behavior variable X3 as the independent variable X; Treat each variable as a node in a causal graph; Through causal inference theory and data correlation analysis, the causal relationship path between variables is determined, and the recommended content variable X1 directly affects the final result Y of user behavior; User context variables X2 and user historical behavior variables X3 indirectly affect the final result Y of user behavior by adjusting preferences for recommended content; Construct a cause-effect diagram to clarify the cause-effect path between variables.

4. The service recommendation quality evaluation method according to claim 3, characterized in that: According to the causal relationship path, the causal effect analysis is performed on the variables on the causal path, and the preliminary causal effect value is output. The specific steps are as follows: According to the causal relationship path in the causal diagram, the independent variables are divided into direct causal variables that directly affect the dependent variable Y, indirect causal variables that indirectly affect the dependent variable Y through direct causal variables, and irrelevant variables that do not appear in the causal relationship path; Keep direct and indirect causal variables and eliminate irrelevant variables; According to the values ​​of the key variables in the independent variables, the recommended content variable X1, the user context variable X2 and the user historical behavior variable X3 are grouped, with obvious differences between the groups, while maintaining the consistency of other independent variables; The group with high values ​​of each key variable in the independent variable is used as the experimental group, and the group with low values ​​of the key variable is used as the control group; Calculate the mean difference of the dependent variable Y between the experimental group and the control group; The group difference value of each key variable is taken as the preliminary causal effect value of the corresponding key causal variable.

5. The service recommendation quality evaluation method according to claim 4, characterized in that: The user behavior causal inference model is constructed by combining causal inference theory and real-time feedback mechanism. The specific steps are as follows: Based on the logistic regression model in causal inference theory, the probability of users receiving recommended content is calculated, and a user behavior causal inference model is constructed. The expression is as follows: ； in, is the probability of the user receiving the recommended content variable X1, is the intercept term, is the coefficient of the user context variable X2, is the coefficient of the user's historical behavior variable X3; Mix the user context variable X2, the user history behavior variable X3, and the label of whether the user accepts the recommended content variable X1, and divide them into a training set and a validation set; Input the training set into the user behavior causal inference model for training; The log-likelihood function with L2 regularization term introduced by maximizing , the expression is as follows: ； in, is the total number of users, is the index of the number of users, For the The probability that a user accepts the recommendation, For the A label of whether to accept the recommended content variable X1, is the regularization coefficient, ; During the iterative training of the user causal inference model, the gradient direction is updated , get the best ; When the maximum number of iterations is reached, the training is terminated and the validation set is input to verify the performance of the user behavior causal inference model by calculating the accuracy, recall rate, and F1 score of the user behavior causal inference model. Set the probability threshold, when When it is greater than the probability threshold, the user is considered to have accepted the recommendation, otherwise, the user is considered to have not accepted the recommendation; Based on the probability calculation results of users receiving recommended content, users who accepted the recommendation are used as the experimental group, and users who did not accept the recommendation are used as the control group; The nearest neighbor matching method is used to match users according to the Euclidean distance of the probability of users receiving recommended content, and the experimental group users are matched with the control group users whose probability of users receiving recommended content is closest to that of users in the experimental group. After the matching is completed, the mean difference of the dependent variable Y in the experimental group and the control group is calculated as the causal effect value of the recommended content on the user's decision; Collect the latest user behavior data in real time, recalculate the probability of the user receiving the recommended content variable X1, and dynamically adjust the probability threshold based on the latest probability of the user receiving the recommended content.

6. The service recommendation quality evaluation method according to claim 5, characterized in that: The method is based on the user behavior causal inference model, and the users are grouped individually, and the causal weights of the users for different recommendation factors are calculated. The specific steps are as follows: The recommended content variable X1, user context variable X2 and user historical behavior variable X3 are used as key features for clustering; By plotting the change curve of the sum of squared clustering errors (SSE) with the number of clusters k, the k value at the inflection point is selected as the optimal k value; Use K-Means algorithm to cluster users and divide them into k different clusters; Input the recommended content variable X1, user context variable X2 and user historical behavior variable X3 of each cluster into the user behavior causal inference model to obtain the probability of users in each cluster receiving the recommended content; The boundary effect values ​​in the user context variable X2 and the user historical behavior variable X3 are calculated as personalized weights to express the influence of user behavior in the cluster. The expression is as follows: ； in, is the personalized weight of the user context variable X2 and the user historical behavior variable X3, ; Normalize the personalization weights.

7. The service recommendation quality evaluation method according to claim 6, characterized in that: The specific steps of dynamically adjusting the recommendation strategy based on the personalized grouping and causal weight of the user are as follows: Collect user context data and user behavior data in real time through a unified data access interface; Update the personalized weights of the user context variable X2 and the user historical behavior variable X3 based on the real-time user context data and user behavior data; For each service item, the product of the updated personalized weight and the user context variable X2 and the user historical behavior variable X3 is calculated as the comprehensive score of the service item; Optimize recommendation strategies according to different types of users.

8. A service recommendation quality evaluation system, based on the service recommendation quality evaluation method according to any one of claims 1 to 7, characterized in that: Including data collection module, causal relationship path module, causal effect analysis module, model building module, causal weight calculation module and recommendation strategy adjustment module; The data collection module is used to collect and process the multidimensional data of users and service resources in the recommendation scenario and output a time-series multidimensional data set; The causal relationship path module is used to define the causal relationship variables of the service recommendation quality evaluation based on the multidimensional data set, and to clarify the causal relationship path between the variables by using the causal inference theory; The causal effect analysis module is used to perform causal effect analysis on the variables on the causal path according to the causal relationship path, and output preliminary causal effects; The model building module is used to build a user behavior causal inference model by combining causal inference theory and real-time feedback mechanism; The causal weight calculation module is used to perform personalized grouping of users based on the user behavior causal inference model and calculate the causal weights of users for different recommendation factors; The recommendation strategy adjustment module is used to dynamically adjust the recommendation strategy based on the personalized grouping and causal weight of the user.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the service recommendation quality evaluation method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the service recommendation quality evaluation method described in any one of claims 1 to 7 are implemented.

Images