CN108334575B - Recommendation result sorting correction method and device and electronic equipment - Google Patents
Recommendation result sorting correction method and device and electronic equipment Download PDFInfo
- Publication number
- CN108334575B CN108334575B CN201810064755.6A CN201810064755A CN108334575B CN 108334575 B CN108334575 B CN 108334575B CN 201810064755 A CN201810064755 A CN 201810064755A CN 108334575 B CN108334575 B CN 108334575B
- Authority
- CN
- China
- Prior art keywords
- recommendation
- recommendation result
- probability distribution
- current
- distribution model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to a recommendation result sorting correction method, belongs to the technical field of computers, and solves the problem that in the prior art, the recommendation result accuracy is low in a recommendation sorting method. The recommendation result ranking correction method disclosed by the embodiment of the application comprises the following steps: the method comprises the steps of obtaining real-time behavior data of a current user on historical recommendation results, constructing a probability distribution model of the current user for executing preset behaviors aiming at the current recommendation results according to the real-time behavior data, and further correcting the current recommendation results according to sampling values of the probability distribution model. According to the ranking correction method, ranking of the recommendation result list is corrected based on the real-time behavior statistics of the user, the recommendation result is obtained based on the behavior habits of the user, the accuracy of the recommendation result is effectively improved, and meanwhile user experience is improved.
Description
Technical Field
The present application relates to the field of computer technologies, and in particular, to a recommendation result ranking correction method and apparatus, and an electronic device.
Background
Personalized ranking for users is an important research topic in recommendation systems. With the rapid development of network platform services, hundreds of merchants and products are recalled each time a user accesses a booth on a platform page, and the merchants and the products enter a candidate set to be displayed to the user. In the prior art, the recommendation result list is mainly obtained by a recommendation model, such as a Learn to rank (ranking learning) model. The Learn to rank model applies a machine learning technology to ranking, and the problem is abstracted into an optimization problem through ranking the recommendation result of each user; the data source is a user behavior log in the past period, and the optimization problem is solved by constructing and sequencing features through feature engineering. The sorting method is limited by user behavior logs acquired by a system, and due to the problem of error points of users or log recording on a line, a lot of noise exists in the obtained logs; training a machine learning model based on a behavior log will cause inaccuracy of the learned model, and in order to improve the recommendation effect, the sequence of a recommendation result list is usually corrected.
The commonly used correction methods in the prior art include: manually intervening the recommendation result list, and performing weight reduction treatment on recommendation results which are exposed for many times but not clicked all the time; alternatively, cross features of the user-recommendation dimension are added to the training data. However, when the weight is adjusted based on the click rate, under the condition that the exposure times of the recommendation result are enough, the click rate is a constant, namely, the weight adjustment value of each time of the recommendation result is fixed, and the function of optimizing the recommendation result is not achieved. And the cross feature of the user-recommendation result dimension is difficult to measure, and the sequencing of the recommendation result cannot be effectively corrected.
Therefore, the recommendation result sorting correction method in the prior art still cannot solve the problem of low accuracy of the recommendation result.
Disclosure of Invention
The application provides a recommendation result sorting correction method, which at least solves the problem that the recommendation result accuracy rate is low in a recommendation sorting method in the prior art.
In order to solve the above problem, in a first aspect, an embodiment of the present application provides a recommendation result ranking correction method, including:
acquiring real-time behavior data of a current user on a historical recommendation result;
according to the real-time behavior data, constructing a probability distribution model of the current user for executing a preset behavior aiming at the current recommendation result;
and correcting the current recommendation result according to the sampling value of the probability distribution model.
In a second aspect, an embodiment of the present application provides a recommendation result ranking correction apparatus, including:
the real-time behavior data acquisition module is used for acquiring the real-time behavior data of the current user on the historical recommendation result;
the user behavior probability distribution model construction module is used for constructing a probability distribution model of the current user for executing the preset behavior aiming at the current recommendation result according to the real-time behavior data;
and the sequencing correction module is used for correcting the current recommendation result according to the sampling value of the probability distribution model.
In a third aspect, an embodiment of the present application further discloses an electronic device, which includes a memory, a processor, and a computer program that is stored in the memory and is executable on the processor, and when the processor executes the computer program, the recommendation result ranking correction method according to the embodiment of the present application is implemented.
In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, where the computer program is executed by a processor to perform the steps of the recommendation result ranking modification method disclosed in the present application.
According to the recommendation result ranking correction method disclosed by the embodiment of the application, the problem of low recommendation result accuracy of a recommendation ranking method in the prior art is solved by acquiring real-time behavior data of a current user on a historical recommendation result, then constructing a probability distribution model of the current user for executing preset behaviors on the current recommendation result according to the real-time behavior data, and further correcting the current recommendation result according to a sampling value of the probability distribution model. According to the method and the device, the ranking of the recommendation result list is corrected based on the real-time behavior statistic of the user, the recommendation result is obtained based on the behavior habit of the user, the accuracy of the recommendation result is effectively improved, and meanwhile, the user experience is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
Fig. 1 is a flowchart of a recommendation result ranking correction method according to a first embodiment of the present application;
FIG. 2 is a schematic view of a Beta distribution constructed by the method for modifying a recommended result ranking according to the second embodiment of the present application;
fig. 3 is a schematic structural diagram of a recommendation result ranking correction apparatus according to a third embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Example one
As shown in fig. 1, the method for modifying recommendation result ranking disclosed in this embodiment includes: step 110 to step 130.
And step 110, acquiring real-time behavior data of the current user on the historical recommendation result.
In specific implementation, a preset behavior list of a current user for a displayed recommendation result is analyzed from a real-time behavior data stream of the user in the whole network, then statistical analysis is performed on the behavior list of the current user from the category dimension of the recommendation result, and the click times and the non-click times of the user for the exposed (displayed) recommendation results of different categories in a past specified time period (such as two hours) are determined. In specific implementation, the preset behavior is determined according to the optimization target of the recommendation result. For example, for a recommendation system with an optimization goal of improving click rate, the preset behavior includes "click"; for a recommendation system with the goal of improving the rate of access to purchases as an optimization goal, the preset behavior comprises "purchase".
In specific implementation, statistical analysis may be performed on the current user behavior list from the recommendation result dimension to determine the number of clicks and the number of non-clicks of different exposed (i.e., displayed) recommendation results of the user within a specified time period (e.g., two hours) in the past.
And 120, constructing a probability distribution model of the current user executing preset behaviors according to the real-time behavior data and aiming at the current recommendation result.
In specific implementation, the real-time behavior data includes: and the user performs preset action times on the exposed recommended result. Taking a preset behavior as a click as an example, the real-time behavior data includes: the number of clicks and the number of non-clicks of the recommended result of the exposure by the user within a past preset time (e.g., within two hours). And then, constructing a probability distribution model of the user for executing click behaviors on the current recommendation result by taking the click times and the non-click times as parameters. In specific implementation, the probability distribution model may be a Beta distribution model or a gaussian distribution model.
And step 130, correcting the current recommendation result according to the sampling value of the probability distribution model.
After the probability distribution model is constructed, the probability distribution curve is sampled, the obtained sampling value is used as weight, the recommendation score of the recommendation result in the recommendation result list is corrected, and the recommendation result list is further reordered, so that the effect of correcting the recommendation result is achieved.
According to the recommendation result ranking correction method disclosed by the embodiment of the application, the problem of low recommendation result accuracy of a recommendation ranking method in the prior art is solved by acquiring real-time behavior data of a current user on a historical recommendation result, then constructing a probability distribution model of the current user for executing preset behaviors on the current recommendation result according to the real-time behavior data, and further correcting the current recommendation result according to a sampling value of the probability distribution model. According to the method and the device, the ranking of the recommendation result list is corrected based on the real-time behavior statistic of the user, the recommendation result is obtained based on the behavior habit of the user, the accuracy of the recommendation result is effectively improved, and meanwhile, the user experience is improved.
Example two
The embodiment is a specific embodiment of the recommended result sorting and correcting method disclosed in the application.
Based on the first embodiment, in specific implementation, the acquiring of the real-time behavior data of the current user on the historical recommendation result includes: and acquiring real-time behavior data executed by the current user on the three-level category recommendation result of the historical exposure. When the real-time behavior data of the current user on the historical recommendation result is obtained, the real-time behavior data can be obtained from the dimension of the category, and the real-time behavior data can also be obtained from the product dimension. Because the recommended system has a plurality of products, the problem of data sparsity may exist when real-time data is acquired from product dimensions, that is, some products may never be exposed in a preset time period, and the system cannot acquire the real-time data stream of the user to the products. Therefore, it is preferable to acquire real-time behavior data executed by the recommendation result of the current user to the historical exposure based on the category dimension. In specific implementation, the products of the recommendation system can be provided with different grade labels. For example, the recommended result "jinquanhong imax international movie city" is labeled "leisure entertainment", the second class label "leisure", the third class label "movie theater", and the fourth class label "sunny district". Therefore, the primary and secondary categories are too wide and not fine, and real-time behavior data executed by the user on the recommendation results of the primary and secondary categories of the historical exposure cannot reflect the distinguishing degree of the user on different recommendation results. The class of the four levels and the lower levels are too fine, and the problem of data sparseness may exist when real-time data are acquired from the dimensionality of the class of the four levels and the lower levels. Preferably, the method and the device for recommending the three-level categories acquire real-time behavior data executed by the current user on the three-level category recommending result of the historical exposure.
In specific implementation, the real-time behavior data includes: and in a specified time period, executing the times of the preset actions and the times of not executing the preset actions on the recommended results under each three-level category of the historical exposure.
When the method is specifically implemented, the preset behavior is determined according to an optimization objective of a recommendation result, and the preset behavior includes but is not limited to: click or purchase. If the recommendation system takes the click rate of the recommendation result as an optimization target, the preset behaviors comprise: clicking a recommendation result; if the recommendation system takes the visit and purchase rate of the recommendation result to be improved as an optimization target, the preset behaviors comprise: and purchasing the recommendation result.
In the following, a specific technical scheme for acquiring real-time behavior data executed by the current user for the three-level category recommendation result of the historical exposure is described in detail by taking a preset behavior as a click example.
Firstly, all recommendation results exposed to the current user in a specified time period are obtained from the real-time data of the whole network. The specified time period is determined according to specific requirements, and can be the last two hours, for example. And then, classifying the exposed recommendation results according to the three-level category labels of the recommendation results, wherein the recommendation result corresponding to each three-level category comprises a plurality of recommendation results. In specific implementation, the total exposure times of all recommendation results under the category corresponding to each tertiary class are taken as the recommendation of the tertiary classThe total exposure times of the result to the current user; taking the total click times of all recommendation results under the category corresponding to a certain three-level category of the current user as the total click times of the recommendation results of the current user on the three-level category; and subtracting the total number of clicks of the current user on the third-level item recommendation result from the total number of exposures of each third-level item to the current user to obtain the number of exposure clicks of the current user on the third-level item recommendation result. Assume that the recommended results that have been exposed within the last two hours include: c1,C2,C3Three classes, class C1Including the recommendation p1And p2If the result p is recommended1Exposing 50 times, clicking 10 times by the current user, and recommending a result p2Exposing for 50 times to obtain class C1The current user is exposed 100 times in total and clicks 10 times. According to the method, the number of exposure and non-click times and the number of exposure and click times of each three-level category recommendation result which has been exposed by the current user in the specified time period can be obtained.
In specific implementation, modeling can be performed by a machine learning method (such as maximum likelihood), and the real-time behavior data of the current user is combined to be used as training data to perform parameter learning, so as to obtain parameters of the probability distribution model.
And then, according to the acquired real-time behavior data of the current user, constructing a probability distribution model of the current user for executing click behaviors according to the current recommendation result. In specific implementation, according to the real-time behavior data, constructing a probability distribution model for the current user to execute a preset behavior with respect to the current recommendation result includes: and determining model parameters according to the times of executing the preset behaviors and the times of not executing the preset behaviors for the recommendation result of each three-level category of the historical exposure by the current user, and constructing a probability distribution model of executing the preset behaviors for the recommendation result of the corresponding three-level category by the current user.
And during specific implementation, constructing a probability distribution model of the current user for executing click behaviors for the current recommendation result according to the exposure and non-click times and the exposure and click times of each three-level category recommendation result which is exposed by the current user. The probability Distribution model may be a Beta Distribution (Beta Distribution) Distribution model or a gaussian Distribution model. In this embodiment, by taking the construction of a Beta distribution model as an example, a probability distribution model for the current user to execute a click behavior for the current recommendation result is constructed according to the number of exposure clicks and the number of exposure clicks of each recommendation result of the current user for each class of three levels that has been exposed.
The Beta distribution (Beta distribution) model is a density function of conjugate prior distribution as bernoulli distribution and binomial distribution, and has important application in machine learning and mathematical statistics. In probability theory, a Beta distribution refers to a set of continuous probability distributions defined in the (0,1) interval. The Beta distribution model has two parameters, which are α and β, respectively, in this embodiment, α ═ γ + the number of exposure clicks, β ═ γ + the number of exposure non-clicks, where γ is a small constant, for example, a value of 1.
With current user pair having been exposed to a tertiary class C1The number of exposure clicks and the number of exposure clicks of the recommended result, and the current user is constructed aiming at the class C which belongs to the third class in the current recommended result1The recommended result of (2) is to perform the Beta distribution model of the click action, for example, α is 1+10, β is 1+90, and then the Beta distribution curve shown in fig. 2 can be obtained.
In the present embodiment, the mean value E of the Beta distribution is 11/102, and the mean value is small, that is, when the Beta distribution shown in fig. 2 is sampled, the sampled values obtained at a high probability are small, so that the recommendation score is adjusted by using the sampled values of the Beta distribution as the weight, and the function of ranking the tripolar recommendation results corresponding to the Beta distribution is achieved. Therefore, the size of the Beta-distributed sampling value is positively correlated with the recent exposure click rate of the current user on the third-level class recommendation result, and the search requirements of the user are fully embodied.
According to the method, the click behavior probability distribution model of the user for each three-level category recommendation result can be obtained. If the obtained real-time behavior data executed by the current user on the recommendation result of a certain three-level product of historical exposure has more exposure click times, the mean value of Beta distribution is larger, and the sampling value of Beta distribution is more likely to fall near the mean value E, so that the obtained sampling value is larger at this time, the recommendation score is adjusted by taking the sampling value of Beta distribution as the weight, and the effect of improving the ranking of the recommendation results of the three-level products corresponding to the Beta distribution is achieved.
In specific implementation, if the recommendation system takes the promotion of the access rate as an optimization target, a preset behavior probability distribution model of each three-level item recommendation result of the user is constructed according to the acquired real-time purchasing behavior data executed by the current user on the three-level item recommendation results of the historical exposure.
During specific implementation, a gaussian distribution model can be constructed according to the obtained real-time behavior data of the user, which is not described in this embodiment again.
After a preset behavior probability distribution model of each three-level class recommendation result of the user is obtained, correcting a corresponding recommendation result in the current recommendation result according to a sampling value of the probability distribution model.
In specific implementation, the modifying the current recommendation result according to the sampling value of the probability distribution model includes: randomly sampling a probability distribution curve corresponding to the probability distribution model to determine a sampling value of the probability distribution model; taking the product of the sampling value and the recommendation score of the current recommendation result as the recommendation score after the current recommendation result is corrected; and correcting the sequence of the current recommendation result according to the corrected recommendation score. And after a preset behavior probability distribution model of each three-level class recommendation result is constructed, respectively correcting the recommendation scores of the recommendation results belonging to the corresponding three-level classes according to the sampling value of each probability distribution.
The recommendation result list returned by the recommendation system comprises recommendation results p1For example, because the result p is recommended1The third-class label is C1Therefore, by recommending result C for the class of three classes1The probability distribution curve corresponding to the probability distribution model is sampled to obtain a sampling value sample (Beta (C)1) Then, the sampling value and the recommendation system are used for recommending the result p1Estimated recommendation score (p)1) As the current recommendation p1The final recommendation score of. I.e. by the formula ctr _ new (p)1)＝ctr(p1)*sample(Beta(C1) Calculate recommendation p)1The final recommendation score of (2) is corrected for the ranking score calculated by the recommendation system. If the recommendation result list returned by the recommendation system comprises the recommendation result p3To recommend the result p3The third-class label is C2By recommending results C for the three classes2The probability distribution curve corresponding to the probability distribution model is sampled to obtain a sampling value sample (Beta (C)2) And by the formula ctr _ new (p)3)＝ctr(p3)*sample(Beta(C2) Calculate recommendation p)3Wherein ctr (p)3) For the recommendation system to recommend result p3And (4) the estimated recommendation score.
According to the method, the revised recommendation score of each recommendation result in the recommendation result list returned by the recommendation system can be obtained. And finally, reordering the recommendation results according to the corrected recommendation scores, and outputting the recommendation results to the client for display.
According to the scheme, based on the user three-level quality behavior statistics in the user real-time data stream, the Beta distribution sampling is combined, real-time intervention correction is performed on the recommendation scores estimated by the recommendation model in the recommendation system, the display of the recommendation results with large exposure and few clicking times can be greatly suppressed, the user experience is improved, and the number of active users in the day is increased. The recommendation score is corrected by using the Beta distribution to obtain the random value, so that the diversity of recommendation results returned by a recommendation system can be improved.
According to the recommendation result ranking correction method disclosed by the embodiment of the application, the problem of low recommendation result accuracy of a recommendation ranking method in the prior art is solved by acquiring real-time behavior data of a current user on a historical recommendation result, then constructing a probability distribution model of the current user for executing preset behaviors on the current recommendation result according to the real-time behavior data, and further correcting the current recommendation result according to a sampling value of the probability distribution model. According to the method and the device, the ranking of the recommendation result list is corrected based on the real-time behavior statistic of the user, the recommendation result is obtained based on the behavior habit of the user, the accuracy of the recommendation result is effectively improved, and meanwhile, the user experience is improved.
The recommendation scores are interfered on the three-level categories by utilizing Beta distribution, and for the three-level categories with high user click times, the interference is arranged in front of the recommendation result list, so that the exposure chance is increased; for the three-level categories with few user clicks, the three-level categories are arranged behind the recommendation result list after intervention, so that exposure opportunities are reduced, and the accuracy of the recommendation result is improved. Meanwhile, if the real-time behavior data is directly acquired at the dimension of the recommendation result, the recommendation results with a large number of clicks of the user are always exposed in the past period, the user experience is poor, and the exposure times of the recommendation results belonging to one third-class with the recommendation results with a large number of clicks of the user can be increased by using the third-class, so that the improvement of the recommendation diversity is facilitated.
EXAMPLE III
As shown in fig. 3, the recommendation result ranking correction apparatus disclosed in this embodiment includes:
a real-time behavior data obtaining module 310, configured to obtain real-time behavior data of a current user on a historical recommendation result;
a user behavior probability distribution model constructing module 320, configured to construct, according to the real-time behavior data, a probability distribution model for the current user to execute a preset behavior with respect to the current recommendation result;
and the sorting correction module 330 is configured to correct the current recommendation result according to the sampling value of the probability distribution model.
Optionally, the real-time behavior data obtaining module 310 is further configured to:
and acquiring real-time behavior data executed by the current user on the three-level category recommendation result of the historical exposure.
Optionally, the real-time behavior data includes: and in a specified time period, executing the times of the preset actions and the times of not executing the preset actions on the recommended results under each three-level category of the historical exposure.
In specific implementation, parameters of the probability distribution model can be obtained through a statistical method according to the real-time behavior data of the user; the model can also be modeled by a machine learning method (such as maximum likelihood), and the model is used as training data to perform parameter learning by combining with the real-time behavior data of the current user, so as to obtain the parameters of the probability distribution model, which is not limited in the application.
Optionally, the user behavior probability distribution model constructing module 320 is further configured to:
and determining model parameters according to the times of executing the preset behaviors and the times of not executing the preset behaviors for the recommendation result of each three-level category of the historical exposure by the current user, and constructing a probability distribution model of executing the preset behaviors for the recommendation result of the corresponding three-level category by the current user.
In specific implementation, the probability distribution model may be a Beta distribution model or a gaussian distribution model.
Optionally, the sorting correction module 330 is further configured to:
randomly sampling a probability distribution curve corresponding to the probability distribution model to determine a sampling value of the probability distribution model;
taking the product of the sampling value and the recommendation score of the current recommendation result as the recommendation score after the current recommendation result is corrected;
and correcting the sequence of the current recommendation result according to the corrected recommendation score.
According to the scheme, the probability distribution model is built based on the user three-level quality behavior statistics in the user real-time data stream, the recommendation scores pre-estimated by the recommendation model in the recommendation system are subjected to real-time intervention and correction by combining with the sampling values of probability distribution, the recommendation results with large exposure and few clicking times can be greatly suppressed, the user experience is improved, and the number of active users in the day is improved. The recommendation score is corrected by using the probability distribution random sampling value, so that the diversity of recommendation results returned by a recommendation system can be improved.
Optionally, the preset behavior is determined according to an optimization goal of the recommendation result, and the preset behavior includes: click or purchase.
The recommendation result ranking correction device disclosed by the embodiment of the application acquires real-time behavior data of a current user on a historical recommendation result, then constructs a probability distribution model for the current user to execute preset behaviors aiming at the current recommendation result according to the real-time behavior data, and further corrects the current recommendation result according to a sampling value of the probability distribution model, so that the problem of low recommendation result accuracy in the recommendation ranking method in the prior art is solved. According to the method and the device, the ranking of the recommendation result list is corrected based on the real-time behavior statistic of the user, the recommendation result is obtained based on the behavior habit of the user, the accuracy of the recommendation result is effectively improved, and meanwhile, the user experience is improved.
The recommendation scores are interfered on the three-level categories by utilizing Beta distribution, and for the three-level categories with high user click times, the interference is arranged in front of the recommendation result list, so that the exposure chance is increased; for the three-level categories with few user clicks, the three-level categories are arranged behind the recommendation result list after intervention, so that exposure opportunities are reduced, and the accuracy of the recommendation result is improved. Meanwhile, if the real-time behavior data is directly acquired at the dimension of the recommendation result, the recommendation results with a large number of clicks of the user are always exposed in the past period, the user experience is poor, and the exposure times of the recommendation results belonging to one third-class with the recommendation results with a large number of clicks of the user can be increased by using the third-class, so that the improvement of the recommendation diversity is facilitated.
Correspondingly, the application also discloses an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the recommendation result ranking correction method according to the first embodiment and the second embodiment of the application. The electronic device can be a PC, a mobile terminal, a personal digital assistant, a tablet computer and the like.
The present application also discloses a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the steps of the recommendation result ranking correction method as described in the first and second embodiments of the present application.
The embodiments in the present specification are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.
The recommendation result ranking correction method and device provided by the application are introduced in detail, a specific example is applied in the description to explain the principle and the implementation mode of the application, and the description of the embodiment is only used for helping to understand the method and the core idea of the application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Claims (8)
1. A recommendation result ranking modification method is characterized by comprising the following steps:
acquiring real-time behavior data of a current user on a historical recommendation result;
according to the real-time behavior data, constructing a probability distribution model of the current user for executing a preset behavior aiming at the current recommendation result; the probability distribution model comprises a probability distribution model of preset behaviors of the current user aiming at each three-level class recommendation result;
randomly sampling a probability distribution curve corresponding to the probability distribution model to determine a sampling value of the probability distribution model;
taking the product of the sampling value and the recommendation score of the current recommendation result as the recommendation score after the current recommendation result is corrected;
and respectively correcting the recommendation scores of the current recommendation results belonging to the corresponding three grades according to the corrected recommendation scores.
2. The method of claim 1, wherein the step of obtaining real-time behavior data of the current user on the historical recommendation comprises:
and acquiring real-time behavior data executed by the current user on the three-level category recommendation result of the historical exposure.
3. The method of claim 2, wherein the real-time behavioral data comprises: and in a specified time period, executing the times of the preset actions and the times of not executing the preset actions on the recommended results under each three-level category of the historical exposure.
4. The method of claim 3, wherein the step of constructing a probability distribution model of the current user performing a preset action for the current recommendation according to the real-time action data comprises:
and determining model parameters according to the times of executing the preset behaviors and the times of not executing the preset behaviors for the recommendation result of each three-level category of the historical exposure by the current user, and constructing a probability distribution model of executing the preset behaviors for the recommendation result of the corresponding three-level category by the current user.
5. The method according to any one of claims 1 to 4, wherein the preset behavior is determined according to an optimization objective of the recommendation, and the preset behavior comprises: click or purchase.
6. A recommendation result ranking correction apparatus, comprising:
the real-time behavior data acquisition module is used for acquiring the real-time behavior data of the current user on the historical recommendation result;
the user behavior probability distribution model construction module is used for constructing a probability distribution model of the current user for executing the preset behavior aiming at the current recommendation result according to the real-time behavior data; the probability distribution model comprises a probability distribution model of preset behaviors of the current user aiming at each three-level class recommendation result;
the sequencing correction module is used for randomly sampling a probability distribution curve corresponding to the probability distribution model to determine a sampling value of the probability distribution model; taking the product of the sampling value and the recommendation score of the current recommendation result as the recommendation score after the current recommendation result is corrected; and respectively correcting the recommendation scores of the current recommendation results belonging to the corresponding three grades according to the corrected recommendation scores.
7. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the recommendation ranking modification method of any of claims 1 to 5 when executing the computer program.
8. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of modifying a recommendation ranking according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810064755.6A CN108334575B (en) | 2018-01-23 | 2018-01-23 | Recommendation result sorting correction method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810064755.6A CN108334575B (en) | 2018-01-23 | 2018-01-23 | Recommendation result sorting correction method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108334575A CN108334575A (en) | 2018-07-27 |
| CN108334575B true CN108334575B (en) | 2022-04-26 |
Family
ID=62926203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810064755.6A Active CN108334575B (en) | 2018-01-23 | 2018-01-23 | Recommendation result sorting correction method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108334575B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109257648B (en) * | 2018-09-30 | 2021-03-16 | 武汉斗鱼网络科技有限公司 | Method, device, terminal and storage medium for correcting similarity between live broadcasts |
| CN109558544B (en) * | 2018-12-12 | 2021-04-27 | 拉扎斯网络科技（上海）有限公司 | Sorting method and device, server and storage medium |
| CN109597941B (en) * | 2018-12-12 | 2021-04-27 | 拉扎斯网络科技（上海）有限公司 | Sorting method and device, electronic equipment and storage medium |
| CN111382349B (en) * | 2018-12-29 | 2023-08-15 | 广州市百果园网络科技有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN111782927B (en) * | 2019-05-15 | 2024-05-17 | 北京京东尚科信息技术有限公司 | Article recommendation method and device and computer storage medium |
| CN110347781B (en) * | 2019-07-18 | 2023-10-20 | 深圳市雅阅科技有限公司 | Article reverse arrangement method, article recommendation method, device, equipment and storage medium |
| CN111275493B (en) * | 2020-02-10 | 2023-08-22 | 拉扎斯网络科技（上海）有限公司 | Processing method and device of list data, server and nonvolatile storage medium |
| CN111444438B (en) * | 2020-03-24 | 2023-09-01 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining quasi-recall rate of recall strategy |
| CN111612581A (en) * | 2020-05-18 | 2020-09-01 | 深圳市分期乐网络科技有限公司 | Method, device and equipment for recommending articles and storage medium |
| CN112231593B (en) * | 2020-12-15 | 2021-03-12 | 上海朝阳永续信息技术股份有限公司 | Financial information intelligent recommendation system |
| CN113822737A (en) * | 2021-03-30 | 2021-12-21 | 北京沃东天骏信息技术有限公司 | Information pushing method and device, computer system and storage medium |
| CN113190758B (en) * | 2021-05-21 | 2023-01-20 | 聚好看科技股份有限公司 | Server and media asset recommendation method |
| CN114037486A (en) * | 2022-01-07 | 2022-02-11 | 阿里健康科技(中国)有限公司 | Method for determining user appeal, application bearing method and device |
| CN115083442B (en) * | 2022-04-29 | 2023-08-08 | 马上消费金融股份有限公司 | Data processing method, device, electronic equipment and computer readable storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004782A (en) * | 2010-11-25 | 2011-04-06 | 北京搜狗科技发展有限公司 | Search result sequencing method and search result sequencer |
| CN102193999A (en) * | 2011-05-09 | 2011-09-21 | 北京百度网讯科技有限公司 | Method and device for sequencing search results |
| CN105740444A (en) * | 2016-02-02 | 2016-07-06 | 桂林电子科技大学 | User score-based project recommendation method |
| CN107368519A (en) * | 2017-06-05 | 2017-11-21 | 桂林电子科技大学 | A kind of cooperative processing method and system for agreeing with user interest change |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7930303B2 (en) * | 2007-04-30 | 2011-04-19 | Microsoft Corporation | Calculating global importance of documents based on global hitting times |
| CN102169566A (en) * | 2010-02-26 | 2011-08-31 | 国际商业机器公司 | Method and device for creating recommended items in unfamiliar fields |
| CN104794135B (en) * | 2014-01-21 | 2018-06-29 | 阿里巴巴集团控股有限公司 | A kind of method and apparatus being ranked up to search result |
| CN104866474B (en) * | 2014-02-20 | 2018-10-09 | 阿里巴巴集团控股有限公司 | Individuation data searching method and device |
| EP3139284A1 (en) * | 2015-09-01 | 2017-03-08 | Dream It Get IT Limited | Media unit retrieval and related processes |
-
2018
- 2018-01-23 CN CN201810064755.6A patent/CN108334575B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004782A (en) * | 2010-11-25 | 2011-04-06 | 北京搜狗科技发展有限公司 | Search result sequencing method and search result sequencer |
| CN102193999A (en) * | 2011-05-09 | 2011-09-21 | 北京百度网讯科技有限公司 | Method and device for sequencing search results |
| CN105740444A (en) * | 2016-02-02 | 2016-07-06 | 桂林电子科技大学 | User score-based project recommendation method |
| CN107368519A (en) * | 2017-06-05 | 2017-11-21 | 桂林电子科技大学 | A kind of cooperative processing method and system for agreeing with user interest change |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108334575A (en) | 2018-07-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334575B (en) | Recommendation result sorting correction method and device and electronic equipment | |
| CN107423442B (en) | Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment | |
| US11176330B2 (en) | Generating recommendation information | |
| US20210326674A1 (en) | Content recommendation method and apparatus, device, and storage medium | |
| US20220215032A1 (en) | Ai-based recommendation method and apparatus, electronic device, and storage medium | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| US20200356875A1 (en) | Model training | |
| US9355095B2 (en) | Click noise characterization model | |
| CN109242710B (en) | Social network node influence ordering method and system | |
| US20120143790A1 (en) | Relevance of search results determined from user clicks and post-click user behavior obtained from click logs | |
| EP2960849A1 (en) | Method and system for recommending an item to a user | |
| US20220138770A1 (en) | Method and apparatus for analyzing sales conversation based on voice recognition | |
| US20210019635A1 (en) | Group specific decision tree | |
| WO2020135642A1 (en) | Model training method and apparatus employing generative adversarial network | |
| CN115510042A (en) | Power system load data filling method and device based on generation countermeasure network | |
| CN115221396A (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
| CN115471271A (en) | Method and device for attributing advertisements, computer equipment and readable storage medium | |
| CN109190040B (en) | Collaborative evolution-based personalized recommendation method and device | |
| CN110347916B (en) | Cross-scene item recommendation method and device, electronic equipment and storage medium | |
| CN115393100A (en) | Resource recommendation method and device | |
| CN108197982B (en) | Method and device for analyzing correlation between flow and multimedia object | |
| CN114722267A (en) | Information pushing method and device and server | |
| CN111340540A (en) | Monitoring method, recommendation method and device of advertisement recommendation model | |
| Wang | Forecast model of TV show rating based on convolutional neural network | |
| CN116308545B (en) | Advertisement online dynamic pushing method combining user behaviors |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |