CN114385714A

CN114385714A - Feedback-based multi-row-order fusion method, device and equipment and readable storage medium

Info

Publication number: CN114385714A
Application number: CN202210039274.6A
Authority: CN
Inventors: 梁超; 黄霁
Original assignee: Wuhan University WHU
Current assignee: Wuhan University WHU
Priority date: 2022-01-13
Filing date: 2022-01-13
Publication date: 2022-04-22

Abstract

The invention provides a feedback-based multi-sequencing fusion method, a feedback-based multi-sequencing fusion device, a feedback-based multi-sequencing fusion equipment and a readable storage medium. The method comprises the following steps: sequencing a plurality of objects to be sequenced through at least two sequencing models respectively to obtain a sequencing list output by each sequencing model; fusing all the sorted lists by equal weight to obtain a fused sorted list; acquiring feedback information, wherein the feedback information comprises labels of the first K objects in the fusion ordered list; calculating a weight for each sorted list based on the feedback information; and fusing all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list. According to the invention, the weight of each sorted list is determined based on a feedback mechanism, so that the influence of an inaccurate sorted list on a final result is reduced, and the accuracy of a finally obtained new fused sorted list is improved.

Description

Feedback-based multi-row-order fusion method, device and equipment and readable storage medium

Technical Field

The invention relates to the technical field of data processing, in particular to a feedback-based multi-row-order fusion method, device and equipment and a readable storage medium.

Background

In a scene related to ordering a plurality of objects, for example, a scene for ordering the appreciation of a plurality of movies or a scene for ordering the similarity between a plurality of face images and a standard face image, an ordering model is often used to order a plurality of objects.

In the prior art, in order to improve the accuracy of the ranking results, a plurality of objects are often ranked by a plurality of ranking models respectively, and then the plurality of ranking results are fused in an equal weight manner to obtain a fused ranking result. However, in this way, if there is a poor-performance ranking model among the plurality of ranking models, the accuracy of the obtained fused ranking result will be low.

Disclosure of Invention

The invention mainly aims to provide a feedback-based multi-sequencing fusion method, a feedback-based multi-sequencing fusion device, a feedback-based multi-sequencing fusion equipment and a readable storage medium, and aims to solve the technical problem that in the prior art, a plurality of sequencing models have poor performance, so that the accuracy of an obtained fusion sequencing result is low.

In a first aspect, the present invention provides a feedback-based multiple-rank fusion method, where the feedback-based multiple-rank fusion method includes:

sequencing a plurality of objects to be sequenced through at least two sequencing models respectively to obtain a sequencing list output by each sequencing model;

fusing all the sorted lists by equal weight to obtain a fused sorted list;

acquiring feedback information, wherein the feedback information comprises labels of the first K objects in the fusion ordered list;

calculating a weight for each sorted list based on the feedback information;

and fusing all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list.

In a second aspect, the present invention further provides a feedback-based multi-sequence fusion apparatus, including:

the sorting module is used for sorting the objects to be sorted through at least two sorting models respectively to obtain a sorting list output by each sorting model;

the first fusion module is used for fusing all the sorted lists through equal weights to obtain a fused sorted list;

an obtaining module, configured to obtain feedback information, where the feedback information includes tags of first K objects in the fusion ordered list;

a calculating module for calculating a weight of each sorted list based on the feedback information;

and the second fusion module is used for fusing all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list.

In a third aspect, the present invention further provides a feedback-based multi-endian fusion apparatus, which includes a processor, a memory, and a feedback-based multi-endian fusion program stored on the memory and executable by the processor, wherein when the feedback-based multi-endian fusion program is executed by the processor, the steps of the feedback-based multi-endian fusion method described above are implemented.

In a fourth aspect, the present invention further provides a readable storage medium, on which a feedback-based multi-ranking fusion program is stored, wherein the feedback-based multi-ranking fusion program, when executed by a processor, implements the steps of the feedback-based multi-ranking fusion method as described above.

In the invention, a plurality of objects to be sorted are sorted by at least two sorting models respectively to obtain a sorting list output by each sorting model; fusing all the sorted lists by equal weight to obtain a fused sorted list; acquiring feedback information, wherein the feedback information comprises labels of the first K objects in the fusion ordered list; calculating a weight for each sorted list based on the feedback information; and fusing all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list. According to the invention, the weight of each sorted list is determined based on a feedback mechanism, so that the influence of an inaccurate sorted list on a final result is reduced, and the accuracy of a finally obtained new fused sorted list is improved.

Drawings

Fig. 1 is a schematic hardware structure diagram of a multiple-sequencing fusion device based on feedback according to an embodiment of the present invention;

FIG. 2 is a schematic flow chart illustrating an embodiment of a feedback-based multi-sequencing fusion method according to the present invention;

fig. 3 is a functional block diagram of a multi-sequencing fusion apparatus based on feedback according to an embodiment of the present invention.

The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.

Detailed Description

It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

In a first aspect, an embodiment of the present invention provides a feedback-based multi-queue fusion device, where the feedback-based multi-queue fusion device may be a device with a data processing function, such as a Personal Computer (PC), a notebook computer, or a server.

Referring to fig. 1, fig. 1 is a schematic diagram of a hardware structure of a multiple-row sequential fusion device based on feedback according to an embodiment of the present invention. In this embodiment of the present invention, the multiple-sequencing fusion device based on feedback may include a processor 1001 (e.g., a Central Processing Unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used for realizing connection communication among the components; the user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard); the network interface 1004 may optionally include a standard wired interface, a WIreless interface (e.g., a WI-FI interface, WI-FI interface); the memory 1005 may be a Random Access Memory (RAM) or a non-volatile memory (non-volatile memory), such as a magnetic disk memory, and the memory 1005 may optionally be a storage device independent of the processor 1001. Those skilled in the art will appreciate that the hardware configuration depicted in FIG. 1 is not intended to be limiting of the present invention, and may include more or less components than those shown, or some components in combination, or a different arrangement of components.

With continued reference to FIG. 1, the memory 1005 of FIG. 1, which is one type of computer storage medium, may include an operating system, a network communication module, a user interface module, and a feedback-based multi-tiered fusion program. The processor 1001 may call a multi-queue fusion program based on feedback stored in the memory 1005, and execute the multi-queue fusion method based on feedback provided by the embodiment of the present invention.

In a second aspect, an embodiment of the present invention provides a feedback-based multiple-rank fusion method.

In an embodiment, referring to fig. 2, fig. 2 is a flowchart illustrating an embodiment of a feedback-based multi-sequencing fusion method according to the present invention. As shown in fig. 2, the feedback-based multi-rank fusion method includes:

s10, sequencing a plurality of objects to be sequenced through at least two sequencing models respectively to obtain a sequencing list output by each sequencing model;

in this embodiment, the plurality of objects to be sorted may be a plurality of movies or a plurality of face images. It should be noted that, this is only an illustrative example, and does not constitute a limitation to several objects to be sorted.

Taking a plurality of objects to be sorted as a plurality of movies as an example, scoring each movie through the sorting model 1, and then sorting the plurality of movies from high to low based on a scoring result to obtain a sorted list 1; similarly, each movie is scored through the ranking model 2, and then the movies are ranked in order from high to low based on the scoring result, thereby obtaining a ranked list 2.

Taking a plurality of objects to be sorted as a plurality of face images as an example, scoring the similarity of each face image and a standard face image through a sorting model 1, and then sorting the plurality of face images from high to low based on a scoring result to obtain a sorting list 1; similarly, the similarity between each facial image and the standard facial image is scored through the ranking model 2, and then a plurality of facial images are ranked from high to low based on the scoring result, so that the ranking list 2 is obtained.

It will be readily appreciated that the number of ranking models is equal to the number of ranking lists.

Step S20, fusing all the sorted lists through equal weight to obtain a fused sorted list;

in the embodiment, a plurality of objects to be sorted comprise face images 1-10, and the sorted list comprises a sorted list 1, a sorted list 2 and a sorted list 3. Referring to table 1, table 1 is a schematic table of ordered table 1:

table 1 referring to table 2, table 2 is a schematic table of the ordered list 2;

face image 1	9.5
		Face image 7	9.1
Face image 2	8.8
		Face image 9	8.7
Face image 8	8.2
		Face image 5	7.7
Face image 4	7.3
		Face image 6	6.8
Face image 3	6.5
		Face image 10	6.3

Table 2 referring to table 3, table 3 is a schematic table of ordered table 3:

face image 3	0.96
		Face image 5	0.91
Face image 1	0.88
		Face image 7	0.85
Face image 10	0.83
		Face image 8	0.77
Face image 9	0.75
		Face image 2	0.66
Face image 6	0.63
		Face image 4	0.65

TABLE 3

As shown in tables 1 to 3, since the algorithms adopted by different ranking models are different, the obtained ranking tables are different, and due to the difference of the scoring mechanisms, the corresponding scores of different ranking tables are different, so that each ranking table needs to be normalized, that is, the score of each face image in each ranking table is changed into a value between 0 and 1. For example, after the normalization process, the score of the face image 1 in the sorted list 1 becomes 0.98; the score of the face image 1 in the sorted list 2 becomes 0.95; the score of the face image 1 in the sorted list 3 becomes 0.88. By adopting an equal weight fusion mode, the total score of the face image 1 can be obtained to be 0.98+0.95+ 0.88. By parity of reasoning, the total scores of the face images 1-10 can be obtained, and the face images 1-10 are sorted again according to the total scores of the face images 1-10 to obtain a fusion sorting list. Referring to table 4, table 4 is a schematic table of the fused ordered list.

Face image 1
	Face image 5
Face image 3
	Face image 7
Face image 8
	Face image 6
Face image 10
	Face image 9
Face image 2
	Face image 4

TABLE 4

Step S30, obtaining feedback information, wherein the feedback information comprises labels of the first K objects in the fusion ordered list;

in this embodiment, feedback information is obtained, where the feedback information includes tags of the first K objects in the fusion ordered list. The value of K is set according to actual needs, for example, according to the number of objects to be sorted.

When the number of the objects to be sorted is not large, the label of each object to be sorted can be predetermined and stored; when the number of the objects to be sorted is large, the first K objects in the fusion sorted list can be output to be marked by a user, so that feedback information based on user operation, namely labels of the first K objects in the fusion sorted list, can be obtained.

Taking table 4 as an example, the value of K is 5, that is, the tags of the first 5 objects in the fusion ordered list are obtained. When the label is a positive label (marked by '1'), the object and the standard face image belong to the same face; when the label is a negative label (identified by "0"), it indicates that the object does not belong to the same face as the standard face image.

Step S40, calculating the weight of each ordered list based on the feedback information;

in this embodiment, the feedback information includes labels of the face image 1, the face image 5, the face image 3, the face image 7, and the face image 8, where the labels of the face image 1, the face image 3, and the face image 7 are positive labels, and the labels of the face image 5 and the face image 8 are negative labels.

Further, in one embodiment, step S40 includes:

determining the number of positive labels in the first K objects to be sorted in each sorting list according to the labels of the first K objects; calculating to obtain the score of each sorting list according to the number of positive labels in the first K objects to be sorted of each sorting list; and determining the weight of each sorted list according to the scores of all the sorted lists.

In this embodiment, if K takes 5, the feedback information includes face image 1, face image 5, face image 3, face image 7 and face image 8, where the face image 1, face image 3 and face image 7 are positive labels, and the face image 5 and face image 8 are negative labels, and in combination with tables 1 to 3, it can be determined that the number of labels as positive labels in the top K (K takes 5) objects to be sorted in the sorted list 1 is 3, the number of labels as positive labels in the top 5 objects to be sorted in the sorted list 2 is 2, and the number of labels as positive labels in the top 5 objects to be sorted in the sorted list 3 is 3.

Assume that the scoring rules of the sorted list are: and (3) the number m of positive labels in the first K objects to be sorted in the sorted list is the score of the sorted list. Thus, a score of 3 in ranked list 1, a score of 2 in ranked list 2, and a score of 3 in ranked list 3 may be obtained. On this basis, it can be determined that the weight of ranked list 1 is 3/8, the weight of ranked list 2 is 2/8, and the weight of ranked list 3 is 3/8.

Further, in one embodiment, step S40 includes:

determining the target object to be sorted with the label as a positive label according to the labels of the first K objects; acquiring a score set of a target object to be sorted in each sorted list, calculating to obtain a standard deviation of the score set, and taking the reciprocal of the standard deviation as a score of each sorted list; and determining the weight of each sorted list according to the scores of all the sorted lists.

In this embodiment, if K is 5, the feedback information includes labels of the face image 1, the face image 5, the face image 3, the face image 7, and the face image 8, where the labels of the face image 1, the face image 3, and the face image 7 are positive labels, and the labels of the face image 5 and the face image 8 are negative labels, and the target object to be sorted whose label is a positive label includes the face image 1, the face image 3, and the face image 7.

With reference to table 1, the score sets of the face image 1, the face image 3, and the face image 7 obtained in the sorted table 1 are: 98. 96 and 80, calculating the standard deviation of the score set, and then taking the reciprocal of the standard deviation as the score of the sorted list 1; by analogy, the score of each ordered list can be obtained. On the basis, the weight of each sorted list can be determined according to the scores of all sorted lists.

And step S50, fusing all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list.

In this embodiment, all the sorted lists are not merged based on the equal weight any more, but all the sorted lists are merged based on the weight of each sorted list calculated in step S40, so as to obtain a new merged sorted list.

Further, in one embodiment, step S50 includes:

carrying out normalization processing on the score of each object to be sorted in each sorted list to obtain a new sorted list; multiplying the score of each object to be sorted in each new sorted list by the weight of each sorted list to obtain a sub-score of each object to be sorted in each new sorted list; adding all the sub-scores corresponding to each object to be sorted to obtain a comprehensive score of each object to be sorted; and sequencing all the objects to be sequenced based on the comprehensive scores of all the objects to be sequenced to obtain a new fusion sequencing list.

In this embodiment, the score of each object to be sorted in each sorted list is first normalized, and the specific process may refer to the embodiment of step S20, which is not described herein again.

For example, after normalization, the sub-score of the face image 1 in the sorted list 1 is 0.98; the sub-score of the face image 1 in the sorted list 2 is 0.95; the sub-score of the face image 1 in the sorted list 3 is 0.88; based on the calculation of step S40, it is determined that the weight of sorted list 1 is 3/8, the weight of sorted list 2 is 2/8, and the weight of sorted list 3 is 3/8. Then

By analogy, the comprehensive score of each object to be sorted (the face images 1-10) can be obtained, so that all the objects to be sorted are sorted based on the comprehensive scores of all the objects to be sorted, and a new fusion sorting list is obtained.

In this embodiment, a plurality of objects to be sorted are sorted by at least two sorting models respectively to obtain a sorting list output by each sorting model; fusing all the sorted lists by equal weight to obtain a fused sorted list; acquiring feedback information, wherein the feedback information comprises labels of the first K objects in the fusion ordered list; calculating a weight for each sorted list based on the feedback information; and fusing all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list. According to the embodiment, the weight of each sorted list is determined based on a feedback mechanism, so that the influence of the inaccurate sorted list on the final result is reduced, and the accuracy of the finally obtained new fused sorted list is improved.

Further, in an embodiment, after step S50, the method further includes:

if new feedback information is received, the new feedback information comprises labels of the first L objects in the fusion ordered list, wherein L is larger than K;

and taking the new feedback information as feedback information, and executing the step of calculating the weight of each ordered list based on the feedback information.

In this embodiment, after obtaining the new fusion ordered list, if new feedback information is received, the new feedback information includes tags of the first L objects in the fusion ordered list, where L is greater than K; the new feedback information is used as the feedback information, and the process returns to the step S40 to the step S50.

It is easy to understand that, if the obtained new fused ordered list is not desirable in the case that the feedback information includes the tags of the first K objects in the fused ordered list, the feedback information further includes the tags of the first L objects in the fused ordered list, thereby further improving the accuracy of the finally obtained fused ordered list.

In a third aspect, an embodiment of the present invention further provides a feedback-based multiple-order fusion apparatus.

In an embodiment, referring to fig. 3, fig. 3 is a functional module diagram of a multi-sequencing fusion device based on feedback according to an embodiment of the present invention. As shown in fig. 3, the feedback-based multiple-rank fusion apparatus includes:

the sorting module 10 is configured to sort the plurality of objects to be sorted through at least two sorting models respectively to obtain a sorted list output by each sorting model;

the first fusing module 20 is configured to fuse all the sorted lists by equal weight to obtain a fused sorted list;

an obtaining module 30, configured to obtain feedback information, where the feedback information includes tags of first K objects in the fusion ordered list;

a calculating module 40 for calculating a weight of each sorted list based on the feedback information;

and a second fusing module 50, configured to fuse all the sorted lists based on the weight of each sorted list to obtain a new fused sorted list.

Further, in an embodiment, the calculating module 40 is configured to:

determining the number of positive labels in the first K objects to be sorted in each sorting list according to the labels of the first K objects;

calculating to obtain the score of each sorting list according to the number of positive labels in the first K objects to be sorted of each sorting list;

and determining the weight of each sorted list according to the scores of all the sorted lists.

Further, in an embodiment, the calculating module 40 is configured to:

determining the target object to be sorted with the label as a positive label according to the labels of the first K objects;

acquiring a score set of a target object to be sorted in each sorted list, calculating to obtain a standard deviation of the score set, and taking the reciprocal of the standard deviation as a score of each sorted list;

Further, in an embodiment, the second fusion module 50 is configured to:

carrying out normalization processing on the score of each object to be sorted in each sorted list to obtain a new sorted list;

multiplying the score of each object to be sorted in each new sorted list by the weight of each sorted list to obtain a sub-score of each object to be sorted in each new sorted list;

adding all the sub-scores corresponding to each object to be sorted to obtain a comprehensive score of each object to be sorted;

and sequencing all the objects to be sequenced based on the comprehensive scores of all the objects to be sequenced to obtain a new fusion sequencing list.

Further, in an embodiment, the feedback-based multi-sequencing fusion apparatus further includes a loop module configured to:

The function implementation of each module in the feedback-based multi-row sequence fusion device corresponds to each step in the feedback-based multi-row sequence fusion method embodiment, and the function and implementation process thereof are not described in detail herein.

In a fourth aspect, the embodiment of the present invention further provides a readable storage medium.

The readable storage medium of the present invention stores a feedback-based multi-queue fusion program, wherein the feedback-based multi-queue fusion program, when executed by a processor, implements the steps of the feedback-based multi-queue fusion method as described above.

The method implemented when the feedback-based multi-sequencing fusion program is executed may refer to each embodiment of the feedback-based multi-sequencing fusion method of the present invention, and details thereof are not repeated herein.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for causing a terminal device to execute the method according to the embodiments of the present invention.

The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims

1. A feedback-based multi-sequencing fusion method is characterized by comprising the following steps:

fusing all the sorted lists by equal weight to obtain a fused sorted list;

calculating a weight for each sorted list based on the feedback information;

2. The feedback-based multi-rank fusion method of claim 1, wherein the step of calculating a weight for each rank list based on the feedback information comprises:

3. The feedback-based multi-rank fusion method of claim 1, wherein the step of calculating a weight for each rank list based on the feedback information comprises:

4. The feedback-based multi-rank fusion method of claim 1, wherein the step of fusing all the ranked lists based on the weight of each ranked list to obtain a new fused ranked list comprises:

5. The feedback-based multi-rank fusion method of claim 1, further comprising, after the step of fusing all ranked lists based on the weight of each ranked list to obtain a new fused ranked list:

6. A feedback-based multi-row order fusion apparatus, comprising:

7. The feedback-based multi-sequencing fusion device of claim 6, wherein the calculation module is configured to:

8. The feedback-based multi-sequencing fusion device of claim 6, wherein the calculation module is configured to:

9. A feedback-based multi-out-of-order fusion device comprising a processor, a memory, and a feedback-based multi-out-of-order fusion program stored on the memory and executable by the processor, wherein the feedback-based multi-out-of-order fusion program when executed by the processor implements the steps of the feedback-based multi-out-of-order fusion method of any of claims 1 to 5.

10. A readable storage medium having stored thereon a feedback-based multi-row sequential fusion program, wherein the feedback-based multi-row sequential fusion program when executed by a processor implements the steps of the feedback-based multi-row sequential fusion method of any one of claims 1 to 5.