A data processing system for determining traffic models

Abstract

The present application provides a data processing system for determining a traffic model. A second model with a combined error within a preset range is selected through training, and is put into use simultaneously with the currently used model within a preset observation time period. A reasonable observation time is preset, and the deviation mean corresponding to each model is observed. The model corresponding to the minimum deviation mean within the preset observation time is selected for continued use, and the rest are withdrawn and no longer used. In addition, the present invention can also adjust the traffic distribution law of the third model according to the efficiency of the third model in processing traffic, so that the third model except the currently used model inputs the same traffic data, which can enable users to observe the prediction effect of the model more accurately and intuitively, and then select the most appropriate prediction model, and make it more convenient for users to observe the data and more flexible to set the data.

Description

Technical Field

[0001] The present application relates to the field of model prediction technology, and in particular to a data processing system for determining a traffic model. Background Art

[0002] With the development of technology, model changes are becoming more and more frequent, and users have higher and higher requirements for the accuracy of model predictions. Usually, when technicians replace the existing model, they first stop using the existing model and then replace the new model with the existing model. When the technicians observe that the use of the new model is not ideal, they switch back to the existing model.

[0003] In this case, it is impossible to run the existing usage model and the new model in parallel for comparison, resulting in inaccurate results. Moreover, when problems arise when using the new model, switching back to the existing usage model wastes time and resources, which also results in some losses and cannot guarantee the accuracy of the prediction. Summary of the Invention

[0004] In response to the above technical problems, the technical solutions adopted in this application are:

[0005] A data processing system for determining a traffic model, the system comprising: a first model ID list A, a sample traffic list L, a processor, a database, and a memory storing a computer program, wherein the first model ID list A = (A1, A2, ..., A j ,...,A n ), A j is the jth first model ID, where j ranges from 1 to n, and n is the number of first model IDs. The first model is an untrained traffic model; the sample click rate list L = (L1, L2, ..., L r ,...,L s ), L r is the mean click rate in the rth first time period, where r ranges from 1 to s and s is the number of time periods;

[0006] When the computer program is executed by a processor, the following steps are implemented:

[0007] S100, obtain the second model ID list B based on the sample flow list and A 1 =(B 1 1, B 1 2, ..., B 1 i ,...,B 1 m ), B iis the i-th second model ID, i ranges from 1 to m, m is the number of second model IDs, and the second model is a trained traffic model;

[0008] S200: Insert the currently used model into the second model list to construct a third model ID list C=(C1, C2, ..., C g ,...,C z ), C g The gth third model ID, where g ranges from 1 to z, and z is the number of third model IDs. The third model ID is the traffic model currently in use.

[0009] S300, obtain the traffic list F of the second time period = (F1, F2, ..., F y ,...,F q ), F y is the flow value of the yth second time period, y ranges from 1 to q, and q is the number of second time periods;

[0010] S400, according to F and C, obtain D y =(D y 1, D y 2, ..., D y g ,...,D y z ), D y g is the mean click-through rate predicted by the g-th third model in the y-th second time period;

[0011] S500, according to D y , get D 0 =(D 0 1, D 0 2, ..., D 0 g ,...,D 0 z ), D 0 g Meet the following conditions:

[0012]

[0013] S600, traverse D 0 , from D 0 Get the maximum value D 0 max , and D 0 max The corresponding third model is used as the message model currently in use.

[0014] The present application has at least the following technical effects: a second model with a combined error within a preset range is selected through training, and is put into use simultaneously with the currently used model within a preset observation time period, a reasonable observation time is preset, the deviation mean corresponding to each model is observed, and the model corresponding to the minimum deviation mean within the preset observation time is selected to continue to be used, and the rest are withdrawn and no longer used; thereby, the prediction effect of the new model can be tested without suspending the currently used model, the error can be reduced as much as possible, and the obtained prediction results can be made more accurate, allowing users to observe the prediction effect of the model more accurately and intuitively, and then select the most suitable prediction model.

[0015] In addition, the present invention can also adjust the flow distribution rate of the third model according to the efficiency of the third model in processing flow, so that the third models except the currently used model input the same flow data, which can reduce the error caused by different flow data, and the user can adjust the flow distribution rate in any second time period, making it more convenient for the user to observe the data and more flexible in setting the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 Provided in the embodiments of this application DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0019] It should be noted that the terms "first," "second," and so on, used in this disclosure are used to distinguish similar objects in terms of naming, but these objects themselves are not limited by these terms. It should be understood that these terms are interchangeable where appropriate without departing from the scope of this disclosure. For example, "first object information" could be described as "second object information," and similarly, "second object information" could be described as "first object information."

[0020] Furthermore, the terms "comprises," "includes," "has," and any variations thereof are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to the steps or elements expressly listed but may also include other steps or elements not expressly listed or inherent to the process, method, product, or apparatus.

[0021] The embodiment of the present application provides a data processing system for determining a traffic model, the system comprising: a first model ID list A, a sample traffic list L, a processor, a database, and a memory storing a computer program, wherein the first model ID list A=(A1, A2, ..., A j ,...,A n ), A j is the jth first model ID, j is 1 to n, n is the number of first model IDs, and the first model is the traffic model to be trained; the sample click rate list L = (L1, L2, ..., L r ,...,L s ), L r is the average click rate in the rth first time period, r ranges from 1 to s, and s is the number of time periods.

[0022] Specifically, the first model ID is a unique identifier of the first model. Furthermore, the first model includes but is not limited to: LR model (Logistic Regression Model), fm model (Factorization Machines Model) and tf model (Tensorflow Model); those skilled in the art know that any click-through rate prediction model falls within the protection scope of the present invention.

[0023] Furthermore, the first model is in an offline state during the training process.

[0024] Furthermore, the first time period = the second time period = ... = the rth time period = ... = the sth time period, which can be understood as the time intervals of any time period being the same.

[0025] When the computer program is executed by a processor, Figure 1 As shown, implement the following steps:

[0026] S100, obtain the second model ID list B based on the sample flow list and A 1 =(B 1 1, B 1 2, ..., B 1 i ,...,B 1 m), B i is the i-th second model ID, i ranges from 1 to m, m is the number of second model IDs, and the second model is a traffic model that has passed training.

[0027] Specifically, the second model ID is a unique identifier of the second model.

[0028] Furthermore, B 1 ∈(A1,A2,...,A j ,...,A n ), and the value of j ranges from 1 to n.

[0029] Furthermore, in S100, the following steps are also included:

[0030] S110: Train the first model in A based on L to obtain a click-through rate prediction value list YA r =(YA r 1. YA r 2, ..., YA r j ,...,YA r n ), YA r j A j The predicted click-through rate in the rth time period.

[0031] Specifically, in the embodiment of the present invention, the traffic corresponding to the same message in the rth time period is used to train the first model. It can be understood that when there are more than two messages in the pth time period, the message with the largest traffic value is selected to train the first model, and YA is obtained. r The message with the largest traffic value can be understood as the message displayed the most times in the rth time period.

[0032] Furthermore, in the embodiment of the present invention, a second prediction value list YA may be obtained. r2 =(YA r2 1. YA r2 2, ..., YA r2 j ,...,YA r2 n ), YA r2 j For XA j User conversion rate in the rth time period.

[0033] Specifically, the user conversion rate is the ratio of the number of users who have the maximum predicted probability of clicking on the message published in the rth time period to the number of users who have received the message.

[0034] Specifically, the first model is trained using the same message in the rth time period. It can be understood that when there are more than two messages in the rth time period, the message with the largest traffic value is selected to train the first model, and YA is obtained. r2 .

[0035] S120, according to YA r With L, obtain OE=(OE1, OE2, ..., OE j ,...,OE n ), OE j =(OE 1 j , OE 2 j ,....,OE r j ,...,OE s j ), OE r j for YA r j With L r The ratio in the rth time period.

[0036] S130, according to OE, obtain PE=(PE1, PE2, ..., PE j ,...,PE n ), PE j For OE j The corresponding deviation mean, PE j Meet the following conditions:

[0037]

[0038] S140, according to PE, obtain PE 0 =(PE 0 1. PE 0 2, ..., PE 0 i ,...,PE 0 m ), PE 0 i For PE 0 The mean of the i-th deviation that is smaller than the set threshold.

[0039] Specifically, the PE 0 Initially it is an empty set, and the PE 0 Each value in PE is compared with the set threshold. When the mean deviation in PE is less than or equal to the set threshold, it is stored in PE. 0 , PE 0 i ∈PE.

[0040] Furthermore, in an embodiment of the present invention, the set threshold is 0.01 to 0.1.

[0041] Preferably, the set threshold is 0.01.

[0042] S150, PE-based 0 , obtain the second model ID list B.

[0043] Specifically, OE up i The corresponding first model ID is recorded as B i , and then obtain the second model ID list B.

[0044] S200: Insert the currently used model into the second model list to construct a third model ID list C=(C1, C2, ..., C g ,...,C z ), C g is the gth third model ID, where g ranges from 1 to z, and z is the number of third model IDs. The third model ID is a traffic model used online.

[0045] Specifically, the second model ID is a unique identifier of the second model. Further, z=m+1.

[0046] In this way, there is no need to suspend the use of the currently used model to test the effect of the new model, and the second model can be used in parallel with the currently used model.

[0047] S300, obtain the traffic list F of the second time period = (F1, F2, ..., F y ,...,F q ), F y is the flow value of the yth preset observation time period, where y ranges from 1 to q, and q is the number of the second time period.

[0048] Specifically, in the embodiment of the present invention, those skilled in the art may set q according to needs.

[0049] Furthermore, in an embodiment of the present invention, the unit of the time interval of the second time period is set to days, and the value of q is 1 day. Since the validity period of the model is short, it is necessary to preset a reasonable observation time for the iterative update of the model to make the predicted data more accurate.

[0050] Furthermore, F y =(F y1 , F y2 ,...,F ye ,...,F yhy ), F ye For Fy The flow value of the e-th third time period included in , the value of e ranges from 1 to hy, and hy is the number of the third time period in the y-th day.

[0051] Specifically, in the embodiment of the present invention, those skilled in the art may set hy according to actual needs.

[0052] Preferably, hy=3.

[0053] Preferably, in the embodiment of the present invention, the time interval of any third time period is 1 hour.

[0054] In this way, by selecting several time periods for observation within a day, and the interval time of each time period is the same, the error can be reduced as much as possible, making the obtained prediction results more accurate, so that users can observe the prediction effect of the model more accurately and intuitively.

[0055] Furthermore, S300 also includes the following steps:

[0056] S310, according to C, obtain the initial flow distribution rate list FB = (FB1, FB2, ..., FB g ,...,FB z ), FB g C g The initial flow distribution rate.

[0057] Specifically, those skilled in the art can set FB according to actual needs. g .

[0058] Furthermore, FB1+FB2+…+FB g +……+FB z =1.

[0059] Preferably, FB1>FB2=...=FB g =……=FB z In this way, there is no need to make subsequent traffic allocation rate adjustments, saving users' delivery resources.

[0060] S320, according to FB, obtain X=(X1, X2, ..., X g ,...,X z ), X g C g The efficiency of processing traffic.

[0061] Specifically, X g Meet the following conditions:

[0062]

[0063] Among them, F1 is the flow value of the first second time period, t1 is the second time interval, and the time unit is (day).

[0064] S330: Based on X1, the initial flow distribution ratio is adjusted to obtain a final flow distribution table FC=(FC1, FC2, ..., FC g ,...,FC z ), FC g is the final flow value of the g-th third model.

[0065] Specifically, the final flow value may be understood as the total flow value input into the g-th third model after the end of q second time periods.

[0066] Preferably, FC1>FC2=...=FC g =...=FC z .

[0067] As mentioned above, the third model except the currently used model inputs the same traffic data, which can reduce the error caused by different traffic data, and the user can adjust the traffic allocation rate in any second time period, making it more convenient for the user to observe the data and more flexible in setting the data.

[0068] S400, according to F and C, obtain D y =(D y 1, D y 2, ..., D y g ,...,D y z ), D y g is the mean click-through rate predicted by the g-th third model in the y-th second time period.

[0069] S500, according to D y , get OE y =(OE y 1. OE y 2, ..., OE y g ,...,OE y z ), OE y g It is the ratio of the average real-time click rate in the second time period to the average click rate preset for the g-th third model in the second time period.

[0070] S600, according to OE y , get PE g =(PE1, PE2, ..., PE g ,...,PEz ), PE g For OE y g The corresponding deviation mean, PE g Meet the following conditions:

[0071]

[0072] Through long-term observation and averaging calculations, the judgment data obtained can be made more accurate, allowing technicians to more intuitively see the advantages and disadvantages of the model.

[0073] S700, traversal PE g , get the minimum PE min , and PE min The corresponding third model is used as the message model currently in use.

[0074] Specifically, in the embodiment of the present invention, PE min The corresponding third model is used as the current model. In the third model ID list, except for PE min Models other than the corresponding third model are withdrawn and no longer used.

[0075] The present invention provides a data processing system for determining a traffic model. A second model with a combined error within a preset range is selected through training, and the second model is put into use simultaneously with the currently used model within a preset observation time period. A reasonable observation time is preset, and the deviation mean corresponding to each model is observed. The model corresponding to the minimum deviation mean within the preset observation time is selected for continued use, and the others are withdrawn and no longer used. Therefore, the prediction effect of the new model can be tested without suspending the use of the currently used model, which can minimize errors and make the obtained prediction results more accurate. The user can observe the prediction effect of the model more accurately and intuitively, and then select the most appropriate prediction model.

[0076] In addition, the present invention can also adjust the flow distribution law of the third model according to the efficiency of the third model in processing flow, so that the flow values accumulated by any third model in q second time periods are the same. In this way, the error caused by different flow data can be reduced, and the user can adjust the flow distribution rate in any second time period, making it more convenient for the user to observe the data and more flexible in setting the data.

[0077] Although some specific embodiments of the present application have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present application. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present application. The scope of the present application is defined by the appended claims.

Claims

1. A data processing system for determining a traffic model, characterized in that: The system includes: a first model ID list A, a sample flow list L, a processor, a database and a memory storing a computer program, wherein the first model ID list A=(A1, A2, ..., A j ,...,A n ), A j is the jth first model ID, j is 1 to n, n is the number of first model IDs, and the first model is the traffic model to be trained; the sample click rate list L = (L1, L2, ..., L r ,...,L s ), L r is the mean click rate in the rth first time period, where r ranges from 1 to s and s is the number of time periods; When the computer program is executed by a processor, the following steps are implemented: S100, obtain the second model ID list B based on the sample flow list and A 1 =(B 1 1, B 1 2, ..., B 1 i ,...,B 1 m ), B i is the i-th second model ID, i ranges from 1 to m, m is the number of second model IDs, and the second model is a trained traffic model; S200: Insert the currently used model into the second model list to construct a third model ID list C=(C1, C2, ..., C g ,...,C z ), C g The gth third model ID, where g ranges from 1 to z, and z is the number of third model IDs. The third model ID is the traffic model currently in use. S300, obtain the traffic list F of the second time period = (F1, F2, ..., F y ,...,F q ), F y is the flow value of the yth second time period, y ranges from 1 to q, and q is the number of second time periods; S400, according to F and C, obtain D y =(D y 1, D y 2, ..., D y g ,...,D y z ), D y g is the mean click-through rate predicted by the g-th third model in the y-th second time period; S500, according to D y , get OE y =(OE y 1. OE y 2, ..., OE y g ,...,OE y z ), OE y g is the ratio of the average real-time click rate in the second time period to the average click rate preset for the g-th third model in the second time period; S600, according to OE y , get PE g =(PE1, PE2, ..., PE g ,...,PE z ), PE g For OE y g The corresponding deviation mean, PE g Meet the following conditions: S700, traversal PE g , get the minimum PE min , and PE min The corresponding third model is used as the message model currently in use.

2. The system according to claim 1, wherein: In S100, the following steps are also included: S110: Train the first model in A based on L to obtain a click-through rate prediction value list YA r =(YA r 1. YA r 2, ..., YA r j ,...,YA r n ), YA r j A j The click-through rate predicted in the rth time period; S120, according to YA r With L, obtain OE=(OE1, OE2, ..., OE j ,...,OE n ), OE j =(OE 1 j , OE 2 j ,....,OE r j ,...,OE s j ), OE r j for YA r j With L r The ratio in the rth time period; S130, according to OE, obtain PE=(PE1, PE2, ..., PE j ,...,PE n ), PE j For OE j The corresponding deviation mean, PE j Meet the following conditions: S140, according to PE, obtain PE 0 =(PE 0 1. PE 0 2, ..., PE 0 i ,...,PE 0 m ), PE 0 i For PE 0 The mean of the deviations of the i-th value that is less than the set threshold; S150, PE-based 0 , obtain the second model ID list B.

3. The system according to claim 2, characterized in that The threshold value is set to 0.

01.

4. The system according to claim 1, wherein: After S300, the following steps may also be included: S310, according to C, obtain the initial flow distribution rate list FB = (FB1, FB2, ..., FB g ,...,FB z ), FB g C g Initial flow distribution rate; S320, according to FB, obtain X=(X1, X2, ..., X g ,...,X z ), X g C g The efficiency of handling traffic; S330: Based on X1, the initial flow distribution ratio is adjusted to obtain a final flow distribution table FC=(FC1, FC2, ..., FC g ,...,FC z ), FC g is the final flow value of the g-th third model.

5. The system according to claim 4, characterized in that In S310, FB1+FB2+...+FB g +……+FB z =1.

6. The system according to claim 4, characterized in that In S310, FB1>FB2=...=FB g =……=FB z .

7. The system according to claim 4, wherein: In S320, X g Meet the following conditions: Among them, F1 is the flow value of the first second time period, t1 is the second time interval, and the time unit is (day).

8. The system according to claim 1, wherein: The value of q is 1 day.

9. The system according to claim 1, wherein: F y =(F y1 , F y2 ,...,F ye ,...,F yhy ), F ye For F y The flow value of the e-th third time period included in , the value of e ranges from 1 to hy, and hy is the number of the third time period in the y-th day.

Images