CN117893278A - Material recommendation method and device, computer equipment and storage medium - Google Patents
Material recommendation method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN117893278A CN117893278A CN202311810390.4A CN202311810390A CN117893278A CN 117893278 A CN117893278 A CN 117893278A CN 202311810390 A CN202311810390 A CN 202311810390A CN 117893278 A CN117893278 A CN 117893278A
- Authority
- CN
- China
- Prior art keywords
- materials
- recommended
- task
- sequence
- target
- 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.)
- Pending
Links
- 239000000463 material Substances 0.000 title claims abstract description 386
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000004927 fusion Effects 0.000 claims abstract description 87
- 238000012163 sequencing technique Methods 0.000 claims abstract description 28
- 239000013077 target material Substances 0.000 claims description 47
- 238000010606 normalization Methods 0.000 claims description 39
- 238000004364 calculation method Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 13
- 230000000694 effects Effects 0.000 abstract description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000002093 peripheral effect Effects 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 101100481876 Danio rerio pbk gene Proteins 0.000 description 2
- 101100481878 Mus musculus Pbk gene Proteins 0.000 description 2
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The application relates to a material recommendation method, a device, computer equipment and a storage medium, and relates to the technical field of Internet. The method comprises the following steps: acquiring n predicted values of m materials to be recommended, wherein the n predicted values correspond to n task targets respectively; respectively sequencing m materials to be recommended based on corresponding predicted values by taking each task target as a sequencing basis to obtain n material sequences corresponding to n task targets; calculating fusion scores corresponding to m materials to be recommended respectively based on the material ranks of the m materials to be recommended in the n material sequences respectively; determining recommendation sequences of m materials to be recommended based on fusion scores corresponding to the m materials to be recommended respectively; and recommending materials based on the recommendation sequence. By the method, the multi-task targets can be evaluated in the same dimension, the problems caused by different meanings and larger dimension and distribution differences among different task targets are avoided, the accuracy of material scoring is improved, and the material recommending effect is further improved.
Description
Technical Field
The embodiment of the application relates to the technical field of Internet, in particular to a material recommendation method, a device, computer equipment and a storage medium.
Background
With the rapid development of modern information technology, more and more internet platforms use recommendation systems to recommend materials (such as content information, commodities, etc.) meeting the interests of users to users, and in practical application, in order to improve the satisfaction degree of users on recommended content or maximize the income of the platforms, multiple factors of the recommended content need to be considered.
In the related art, after the predicted values of the materials corresponding to the task targets are obtained, a final score is obtained by fusing the predicted values of the task targets, so as to perform material sorting according to the material sorting, and then material recommendation is performed based on the sorting result.
However, the problems of difficulty in score fusion due to the fact that different meanings and dimension and distribution differences are large exist among different task targets, so that fusion scoring results are inaccurate, material sorting recommendation results are affected, and recommendation effects are poor.
Disclosure of Invention
The embodiment of the application provides a material recommending method, a device, computer equipment and a storage medium, which can avoid the problems caused by different meanings and larger dimension and distribution differences among a plurality of different task targets, improve the accuracy of material scoring and further improve the material recommending effect. The technical scheme is as follows:
in one aspect, a method for recommending materials is provided, the method comprising:
Obtaining n predicted values of m materials to be recommended, wherein the m predicted values are respectively corresponding to n task targets, m is more than or equal to 2, n is more than or equal to 2, and m and n are positive integers;
respectively sequencing the m materials to be recommended based on the corresponding predicted values by taking each task target as a sequencing basis to obtain n material sequences corresponding to the n task targets;
Calculating fusion scores corresponding to the m materials to be recommended respectively based on the material ranks of the m materials to be recommended in the n material sequences respectively;
generating recommendation sequences of the m materials to be recommended based on the fusion scores corresponding to the m materials to be recommended respectively;
and recommending materials based on the recommendation sequence.
In another aspect, there is provided a material recommendation device, the device comprising:
The acquisition module is used for acquiring n predicted values of m materials to be recommended, wherein the m predicted values are respectively corresponding to n task targets, m is more than or equal to 2, n is more than or equal to 2, and m and n are positive integers;
The sorting module is used for sorting the m materials to be recommended based on the corresponding predicted values by taking the task targets as sorting basis to obtain n material sequences corresponding to the n task targets;
The score calculation module is used for calculating fusion scores corresponding to the m materials to be recommended respectively based on the material ranks of the m materials to be recommended in the n material sequences respectively;
the sequence generation module is used for generating recommendation sequences of the m materials to be recommended based on the fusion scores corresponding to the m materials to be recommended respectively;
and the material recommending module is used for recommending materials based on the recommending sequence.
In one possible implementation, the score calculating module includes:
the normalization sub-module is used for respectively carrying out normalization processing on the material ranks in the n material sequences by taking the material sequences as units to obtain n normalization scores of the m materials to be recommended, wherein the n normalization scores correspond to n task targets respectively;
The score calculation sub-module is used for calculating the fusion scores corresponding to the m materials to be recommended respectively based on n normalized scores of the m materials to be recommended respectively.
In one possible implementation manner, the score calculation sub-module is configured to perform score fusion on n normalized scores of a target material based on a target operation rule, so as to obtain a fused score of the target material; the target material is any one of the m materials to be recommended; the target operation rule is used for indicating a score fusion mode and a weighted weight value of a normalized score corresponding to each task target.
In one possible implementation, the score calculation sub-module is configured to.
Acquiring material ranks corresponding to the m materials to be recommended in a target material sequence; the target material sequence is any one of the n material sequences;
Carrying out normalization calculation on the material ranking in the target material sequence based on a normalization formula to obtain normalization scores of the m materials to be recommended in the target material sequence;
The higher the ranking of the materials to be recommended in the material sequence is, the higher the corresponding normalized score is.
In one possible implementation, the material recommendation module is configured to, in use,
Under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the large to the small of the corresponding predicted values, recommending the first K materials to be recommended with the highest fusion score in the recommended sequence;
and recommending the first K materials to be recommended with the lowest fusion score in the recommendation sequence under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the small to the large of the corresponding predicted values.
In one possible implementation, the acquiring module is configured to,
Task information of a target material is obtained; the target material is any one of the m materials to be recommended;
Inputting task information of the target materials into a multi-task learning model to obtain n predicted values, corresponding to the n task targets, of the target materials output by the multi-task learning model;
The multi-task learning model is trained based on task information of a material sample and n task target labels of the material sample corresponding to n task targets.
In one possible implementation, the task information includes at least one of: material information, recommended user information, and recommended scene information.
In another aspect, a computer device is provided, the computer device comprising a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the material recommendation method described above.
In another aspect, a computer readable storage medium having at least one computer program stored therein is provided, the computer program being loaded and executed by a processor to implement the material recommendation method described above.
In another aspect, a computer program product is provided that includes at least one computer program that is loaded and executed by a processor to implement the material recommendation method provided in the various alternative implementations described above.
The technical scheme provided by the application can comprise the following beneficial effects:
According to the material recommending method provided by the embodiment of the application, after a plurality of predicted values of a plurality of materials to be recommended, which correspond to a plurality of task targets, are obtained, the plurality of materials to be recommended are respectively sequenced by taking each task target as a sequencing basis, so as to obtain a material sequence corresponding to each task target, then fusion score calculation is carried out on the basis of the material ranks of the materials to be recommended in the plurality of material sequences, so as to obtain fusion scores of the materials to be recommended, and then a recommending sequence is generated on the basis of the fusion score sequencing of the materials to be recommended, so that material recommendation is carried out according to the recommending sequence; in the process, the fusion score calculation is carried out on the materials to be recommended by converting the numerical value magnitude relation between the predicted values corresponding to the multitask targets into the front-back relation of the material ranking in the material sequence corresponding to the multitask targets, so that the evaluation of the multitask targets is in the same dimension, the problems caused by different meanings and larger dimension and distribution differences among different task targets are avoided, the accuracy of material scoring is improved, and the material recommending effect is further improved.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application as claimed.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.
FIG. 1 illustrates a flow chart of a material recommendation method provided by an exemplary embodiment of the present application;
FIG. 2 illustrates a flow chart of a material recommendation method provided by an exemplary embodiment of the present application;
FIG. 3 illustrates a schematic diagram of a multi-task learning model provided by an exemplary embodiment of the present application;
FIG. 4 illustrates a block diagram of a material recommendation device provided in accordance with an exemplary embodiment of the present application;
FIG. 5 is a block diagram of a computer device shown in accordance with an exemplary embodiment;
fig. 6 is a block diagram of a computer device, according to an example embodiment.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of apparatus and methods consistent with aspects of the application as detailed in the accompanying claims.
The embodiment of the application provides a material recommending method, which can reduce or eliminate the influence of different task targets on the meaning, dimension, distribution difference and the like, synthesizes the predicted values of the materials to be recommended corresponding to a plurality of task targets, improves the score fusion effect, and further improves the material sequencing recommending effect. Fig. 1 shows a flowchart of a material recommendation method according to an exemplary embodiment of the present application, where the method may be performed by a computer device, which may be implemented as a server or a terminal, and as shown in fig. 1, the material recommendation method may include the following steps:
step 110, obtaining n predicted values of m materials to be recommended, wherein the n predicted values respectively correspond to n task targets, m is more than or equal to 2, n is more than or equal to 2, and m and n are positive integers.
The n predicted values of each material to be recommended corresponding to the n task targets can be predicted based on task information of the material to be recommended through a pre-trained recommendation model; the recommendation models applied in the process can be single-task learning models respectively corresponding to the task targets, namely, the recommendation models of the task targets are mutually independent; or the recommendation model applied in the process can also be a multi-task learning model, and the multi-task learning model is used for outputting n predicted values corresponding to n task targets respectively after task information of the materials to be recommended is received.
In the embodiment of the application, the task target is an evaluation index for indicating whether the material to be recommended is worth being recommended, and the larger the predicted value corresponding to the task target is, the more worth being recommended the material to be recommended is indicated in the dimension of the task target; illustratively, the task targets may include CTR (Click-through Rate), interaction (such as sharing, praise, attention, comment, etc.), WATCH TIME (consumption duration), etc., and the types and the number of the task targets may be set differently based on different actual requirements, which is not limited in the present application.
And 120, respectively sequencing m materials to be recommended based on the corresponding predicted values by taking each task target as a sequencing basis, so as to obtain n material sequences corresponding to n task targets.
Because each material to be recommended has the predicted value corresponding to n task targets, when the materials to be recommended are ordered by taking the task targets as the ordering basis, n material sequences corresponding to the n task targets can be obtained, wherein the material ranks of the same material to be recommended in different material sequences can be different based on the difference of the corresponding predicted values.
It should be noted that, in order to ensure that the subsequent fractional fusion can be performed, the ordering modes corresponding to the n material sequences should be kept consistent; the n material sequences are obtained by sequencing the m materials to be recommended from high to low according to the corresponding predicted values, or the n material sequences are obtained by sequencing the m materials to be recommended from low to high according to the corresponding predicted values.
And 130, calculating fusion scores corresponding to the m materials to be recommended respectively based on the material ranks of the m materials to be recommended in the n material sequences respectively.
In order to reduce the fusion problem between the predictive scores of different task targets, in the embodiment of the application, the attention of the score value of the predictive score of the task target is mapped to the attention of the score sorting, so that even if the score difference between the different task targets changes, the dimension or the data distribution difference can be converted into the sorting relation and then can be converted into the same dimension, thereby ensuring the reliability of the score fusion based on the predictive score of the multitask target.
In one possible implementation manner, after obtaining the material ranks of the material to be recommended in the n material sequences, the computer device may directly determine the material ranks of the material to be recommended in the n material sequences as the score of the material to be recommended under the corresponding task target, for example, if the material rank of the material to be recommended 1 in the material sequence corresponding to the task target 1 is 3, the score of the recommended material 1 corresponding to the task target 1 is 3; if the material rank of the material 1 to be recommended in the material sequence corresponding to the task target 2 is 7, the score of the recommended material 1 corresponding to the task target 2 is 7. After the scores of the materials to be recommended corresponding to the task targets are obtained, the fusion scores of the materials to be recommended are calculated based on a preset score fusion mode, for example, the scores of the materials to be recommended corresponding to the task targets are weighted and summed to obtain the fusion scores of the materials to be recommended, and the like.
In another possible implementation manner, after obtaining the material ranks of the materials to be recommended in the n material sequences, the computer device may perform normalization processing on the material ranks of the materials to be recommended in the n material sequences according to a preset normalization rule, so as to calculate the fusion score of the materials to be recommended based on the normalized score of the materials to be recommended corresponding to each task target.
And 140, determining recommendation sequences of the m materials to be recommended based on fusion scores corresponding to the m materials to be recommended.
In the embodiment of the application, the computer equipment can sort the materials to be recommended according to the fusion score of each material to be recommended so as to generate a recommendation sequence containing m materials to be recommended; when the m materials to be recommended are ranked, the computer device may rank according to the order of the fusion score from large to small based on the setting, or may rank according to the order of the fusion score from small to large, which is not limited in the present application.
And step 150, recommending materials based on the recommendation sequence.
In summary, in the material recommending method provided by the embodiment of the present application, after obtaining multiple predicted values of multiple materials to be recommended, which correspond to multiple task targets, the multiple materials to be recommended are sequenced according to the respective task targets, so as to obtain a material sequence corresponding to each task target, then fusion score calculation is performed based on the material ranks of the multiple materials to be recommended in the multiple material sequences, so as to obtain fusion scores corresponding to the respective materials to be recommended, and then a recommending sequence is generated based on the fusion score sequencing of the multiple materials to be recommended, so as to recommend the materials according to the recommending sequence; in the process, the fusion score calculation is carried out on the materials to be recommended by converting the numerical value magnitude relation between the predicted values corresponding to the multitask targets into the front-back relation of the material ranking in the material sequence corresponding to the multitask targets, so that the evaluation of the multitask targets is in the same dimension, the problems caused by different meanings and larger dimension and distribution differences among different task targets are avoided, the accuracy of material scoring is improved, and the material recommending effect is further improved.
Taking a normalized process of material ranks of materials to be recommended in n material ranks according to a preset normalization rule after the material ranks of the materials to be recommended in n material ranks are obtained by a computer device, taking calculating a fusion score of the materials to be recommended as an example based on normalized scores of the materials to be recommended corresponding to respective task targets, fig. 2 shows a flowchart of a material recommendation method provided by an exemplary embodiment of the present application, where the method may be executed by the computer device, and the computer device may be implemented as a server or a terminal, as shown in fig. 2, and the material recommendation method may include the following steps:
step 210, obtaining n predicted values of m materials to be recommended, wherein the n predicted values respectively correspond to n task targets, m is more than or equal to 2, n is more than or equal to 2, and m and n are positive integers.
The computer device may obtain n predicted values of m materials to be recommended corresponding to n task targets through the multi-task learning model, and illustratively, for obtaining n predicted values of one material to be recommended corresponding to n task targets through the multi-task learning model, the process may be implemented as follows:
Task information of a target material is obtained; the target material is any one of m materials to be recommended;
And inputting task information of the target materials into the multi-task learning model to obtain n predicted values of the target materials output by the multi-task learning model, wherein the n predicted values correspond to n task targets.
Wherein the task information may include at least one of: material information, recommended user information and recommended scene information; the recommended user information refers to user information of a recommended user corresponding to the material to be recommended, such as user gender, user age, user geographic position and the like; the recommended scene information is used for indicating the context environment corresponding to the material to be recommended, such as the recommended time, the recommended theme tag (such as entertainment, news and the like); the material information is used to indicate properties of the material, such as a material tag, a material category (e.g., video, graphics, text, links, etc.).
Fig. 3 is a schematic diagram of a multi-task learning model according to an exemplary embodiment of the present application, where the multi-task learning model includes an input layer 310, an expert layer 320, and n task towers 330 corresponding to n task targets, respectively, and task information of a material to be recommended is input into a pre-trained multi-task learning model in a prediction process, so as to obtain n predicted values corresponding to the n task targets output by the n task towers, respectively, as shown in fig. 3.
The multi-task learning model is trained based on task information of a material sample and n task target labels of the material sample corresponding to n task targets; illustratively, the total number of samples is N, N task object labels of the ith material sample corresponding to N task objects are , wherein x i is task information of the material sample i, i is task object label of the material sample i corresponding to the t task object, and t is a value in 1-N; the task prediction function mapping relation corresponding to the multi-task learning model can be expressed as:
ft(x;θs,θt):X→Yt
Wherein, θ s,θt is the shared parameter (including the parameters of the input layer and the expert layer) and the parameter which is exclusive to the specific task (i.e. the parameter corresponding to the task tower) in the multi-task learning model; the total loss function of the multitasking learning model may be defined as follows:
Wherein w t is a weight corresponding to the corresponding task objective t, and the loss function of each task objective t can be defined as follows:
According to the total loss function L total of the multi-task learning model, training, updating and recommending each model parameter in the multi-task learning model by adopting a BP (Backpropagation) error back propagation algorithm, so as to obtain an overall optimal solution model parameter combination of the multi-task learning model, and processing task information of an input target material based on the multi-task learning model to obtain n predicted values of the target material output by the multi-task learning model, wherein the n predicted values correspond to n task targets.
And 220, respectively sequencing the m materials to be recommended based on the corresponding predicted values by taking each task target as a sequencing basis, so as to obtain n material sequences corresponding to n task targets.
Illustratively, taking the example that the task targets include clicking, interaction and viewing consumption time, the number of materials to be recommended is 10, and table 1 shows the predictive scores of the 10 materials to be recommended corresponding to the respective task targets.
TABLE 1
| Item to be recommended | Ctr (click score) | Itr (interaction) | Watchtime (consumption time) |
| item1 | 0.42 | 0.05 | 4.2 |
| item2 | 0.23 | 0.07 | 5.6 |
| item3 | 0.34 | 0.01 | 1.6 |
| item4 | 0.54 | 0.06 | 10.5 |
| item5 | 0.12 | 0.15 | 8.5 |
| itme6 | 0.10 | 0.04 | 9.8 |
| item7 | 0.32 | 0.02 | 6.7 |
| item8 | 0.56 | 0.09 | 11 |
| item9 | 0.15 | 0.11 | 3.5 |
| Item10 | 0.25 | 0.14 | 8.3 |
As can be seen from table 1, the prediction scores corresponding to the task targets have larger differences in the meanings and dimensions of the task targets; after m materials to be recommended are respectively sequenced according to each task target, a material sequence corresponding to each task target is obtained, three material sequences of three task targets shown in table 2, namely a click ranking sequence, an interaction ranking sequence and a consumption duration ranking sequence are obtained by taking sequencing from big to small according to a predicted value as an example, and each number in table 2 represents the material ranking of the corresponding materials to be recommended in the corresponding material sequence.
TABLE 2
And 230, respectively carrying out normalization processing on the material ranks in the n material sequences by taking the material sequences as units, and obtaining n normalization scores of the m materials to be recommended, wherein the n normalization scores correspond to n task targets respectively.
The normalization processing is a processing mode of mapping the value ranges of different features to a unified interval, and can reduce the amplitude difference between the feature values.
Taking the normalization processing for the material rank in the target material sequence in the n material sequences as an example, the process can be implemented as follows:
Acquiring material ranks corresponding to m materials to be recommended in a target material sequence; the target material sequence is any one of n material sequences;
Carrying out normalization calculation on the material ranking in the target material sequence based on a normalization formula to obtain normalization scores of m materials to be recommended corresponding to the target material sequence;
The higher the ranking of the materials to be recommended in the material sequence is, the higher the corresponding normalized score is.
The normalization formula may have different definitions based on different actual processing operations, which is not limited in the embodiment of the present application; the application schematically shows a feasible normalization formula for converting the ranking of materials in each task objective into a score between 0 and 1, wherein the normalization formula is as follows:
Wherein alpha and beta are formula parameters larger than 0, and the higher the ranking of the materials to be recommended is, the higher the corresponding converted score is, so that the sequence relationship obtained based on the normalized score is consistent with the sequence relationship in the material sequence obtained based on the predicted value sequencing; optionally, normalization formulas corresponding to alpha and beta with different sizes can be adopted for normalization processing for different task targets, and the values of the alpha and the beta can be adjusted based on actual conditions.
Illustratively, assuming that the values of the parameters α, β in the normalization formulas of all the task targets are 1, after normalization processing is performed on the material ranks in the three material sequences shown in table 2, the normalized scores shown in table 3 can be obtained:
TABLE 3 Table 3
| Item to be recommended | Rank ctr score | Rank itr min | RANK WATCHTIME min |
| item1 | 1/(1*3+1)＝1/4 | 1/(1*7+1)＝1/8 | 1/(1*8+1)＝1/9 |
| item2 | 1/(1*7+1)＝1/8 | 1/(1*5+1)＝1/6 | 1/(1*7+1)＝1/8 |
| item3 | 1/(1*4+1)＝1/5 | 1/(1*10+1)＝1/11 | 1/(1*10+1)＝1/11 |
| item4 | 1/(1*2+1)＝1/3 | 1/(1*6+1)＝1/7 | 1/(1*2+1)＝1/3 |
| item5 | 1/(1*9+1)＝1/10 | 1/(1*1+1)＝1/2 | 1/(1*4+1)＝1/5 |
| Item6 | 1/(1*10+1)＝1/11 | 1/(1*8+1)＝1/9 | 1/(1*3+1)＝1/4 |
| item7 | 1/(1*5+1)＝1/6 | 1/(1*9+1)＝1/10 | 1/(1*6+1)＝1/7 |
| item8 | 1/(1*1+1)＝1/2 | 1/(1*4+1)＝1/5 | 1/(1*1+1)＝1/2 |
| item9 | 1/(1*8+1)＝1/9 | 1/(1*3+1)＝1/4 | 1/(1*9+1)＝1/10 |
| Item10 | 1/(1*6+1)＝1/7 | 1/(1*2+1)＝1/3 | 1/(1*5+1)＝1/6 |
Step 240, calculating fusion scores corresponding to the m materials to be recommended based on n normalized scores of the m materials to be recommended.
In one possible implementation, the computer device may perform score fusion on n normalized scores of the target material based on the target operation rule, to obtain a fused score of the target material; the target material is any one of m materials to be recommended; the target operation rule is used for indicating a score fusion mode and a weighted weight value of a normalized score corresponding to each task target.
The target operation rule may be addition fusion or multiplication fusion.
The addition fusion refers to weighted summation of normalized scores of a plurality of task targets; illustratively, taking the example of score fusion of the normalized scores corresponding to the target click rate ctr and the viewing duration watchtime, two corresponding weight parameters w1 and w2 are set, and then the addition fusion formula can be expressed as follows:
score=w1×rank ctr division+w2× RANK WATCH TIME division
Further, a nonlinear relationship may be introduced inside the additive fusion to enhance the expressive power of the fusion formula, in which case the additive fusion formula may be expressed as:
Score＝w1*(a1*ctr+b1)c1+w2*(a2*watchtime+b2)c2
multiplication fusion refers to element-by-element multiplication of normalized scores of a plurality of task targets, and the importance degree of the target targets can be adjusted through internal adjustment of target factors; illustratively, taking still the example of score fusion of the normalized scores corresponding to the target click rate ctr and the viewing duration watchtime, the corresponding multiplicative fusion formula can be expressed as:
Score＝(a1*ctr+b1)c1*(a2*watchtime+b2)c2
the target operation rule adopted when calculating the fusion score corresponding to each material to be recommended can be defined based on actual requirements, and the application is not limited to the definition; the following description illustrates a target operation rule, and assuming that the target operation rule is score=rank_ctr+2×rank_itr+rank_watch, the fusion Score of each material to be recommended corresponding to table 3 obtained based on the target operation rule is shown in table 4.
TABLE 4 Table 4
Step 250, generating recommendation sequences of m materials to be recommended based on the fusion scores corresponding to the m materials to be recommended.
The recommendation sequence may be generated according to the order of the fusion score from large to small, or may be generated according to the order of the fusion score from small to large, which is schematically shown in table 4, and if the recommendation sequence is generated according to the order of the fusion score from large to small, the recommendation sequence is: item8-item5-item10-item4-item9-item1-item2-item6-item7-item3; if the recommendation sequence is generated according to the sequence of the fusion score from small to large, the recommendation sequence is as follows: item3-item7-item6-item2-item1-item9-item4-item10-item5-item8.
In step 260, material recommendation is performed based on the recommendation sequence.
In the embodiment of the application, since the sorting mode of the material sequence can influence the material ranking of each material to be recommended in the material sequence, different material recommendation bases can be provided based on different sorting modes of the material sequences, so that the material to be recommended to the corresponding user is the material to be recommended which meets the requirements of the user.
And recommending the first K materials to be recommended with the highest fusion score in the recommendation sequence under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the large to the small of the corresponding predicted values.
In this case, if the recommended sequence is ordered in the order of the fusion score from high to low, the computer device may select to-be-recommended materials of TopK in the recommended sequence to recommend to the user.
If the recommendation sequences are ordered according to the order of the fusion scores from small to large, the computer equipment can select the last K materials to be recommended in the recommendation sequences to recommend the materials to the user.
And recommending the first K materials to be recommended with the lowest fusion score in the recommendation sequence under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the small to the large of the corresponding predicted values.
In this case, if the recommended sequence is ordered in order of the fusion score from small to large, the computer device may select to-be-recommended materials of TopK in the recommended sequence to recommend to the user.
If the recommendation sequences are ordered according to the order of the fusion scores from large to small, the computer equipment can select the last K materials to be recommended in the recommendation sequences to recommend the materials to the user.
In summary, in the material recommending method provided by the embodiment of the present application, after obtaining multiple predicted values of multiple materials to be recommended, which correspond to multiple task targets, the multiple materials to be recommended are sequenced according to the sequencing of the task targets, so as to obtain a material sequence corresponding to the task targets, then the material ranks of the materials to be recommended in the material sequences are normalized, fusion score calculation is performed based on multiple normalization scores corresponding to the materials to be recommended, so as to obtain fusion scores corresponding to the materials to be recommended, and then a recommendation sequence is generated based on the fusion score sequencing of the materials to be recommended, so as to recommend the materials according to the recommendation sequence; in the process, the fusion score calculation is carried out on the materials to be recommended by converting the numerical value magnitude relation between the predicted values corresponding to the multitask targets into the front-back relation of the material ranking in the material sequence corresponding to the multitask targets, so that the evaluation of the multitask targets is in the same dimension, the problems caused by different meanings and larger dimension and distribution differences among different task targets are avoided, the accuracy of material scoring is improved, and the material recommending effect is further improved.
Fig. 4 shows a block diagram of a material recommendation device according to an exemplary embodiment of the present application, which may perform all or part of the steps of the embodiment shown in fig. 1 or fig. 2, and which may include, as shown in fig. 4:
The acquisition module 410 is configured to acquire n predicted values of m materials to be recommended, where m is greater than or equal to 2, n is greater than or equal to 2, and m and n are positive integers;
the sorting module 420 is configured to sort the m materials to be recommended based on the corresponding predicted values by taking the task targets as sorting bases, so as to obtain n material sequences corresponding to the n task targets;
The score calculating module 430 is configured to calculate fusion scores corresponding to the m materials to be recommended respectively, based on the material ranks of the m materials to be recommended in the n material sequences respectively;
The sequence generating module 440 is configured to generate a recommendation sequence of the m materials to be recommended based on the fusion scores corresponding to the m materials to be recommended respectively;
and the material recommending module 450 is used for recommending materials based on the recommending sequence.
In one possible implementation, the score calculation module 430 includes:
the normalization sub-module is used for respectively carrying out normalization processing on the material ranks in the n material sequences by taking the material sequences as units to obtain n normalization scores of the m materials to be recommended, wherein the n normalization scores correspond to n task targets respectively;
The score calculation sub-module is used for calculating the fusion scores corresponding to the m materials to be recommended respectively based on n normalized scores of the m materials to be recommended respectively.
In one possible implementation manner, the score calculation sub-module is configured to perform score fusion on n normalized scores of a target material based on a target operation rule, so as to obtain a fused score of the target material; the target material is any one of the m materials to be recommended; the target operation rule is used for indicating a score fusion mode and a weighted weight value of a normalized score corresponding to each task target.
In one possible implementation, the score calculation sub-module is configured to.
Acquiring material ranks corresponding to the m materials to be recommended in a target material sequence; the target material sequence is any one of the n material sequences;
Carrying out normalization calculation on the material ranking in the target material sequence based on a normalization formula to obtain normalization scores of the m materials to be recommended in the target material sequence;
The higher the ranking of the materials to be recommended in the material sequence is, the higher the corresponding normalized score is.
In one possible implementation, the material recommendation module 450 is configured to,
Under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the large to the small of the corresponding predicted values, recommending the first K materials to be recommended with the highest fusion score in the recommended sequence;
and recommending the first K materials to be recommended with the lowest fusion score in the recommendation sequence under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the small to the large of the corresponding predicted values.
In one possible implementation, the acquiring module 410 is configured to,
Task information of a target material is obtained; the target material is any one of the m materials to be recommended;
Inputting task information of the target materials into a multi-task learning model to obtain n predicted values, corresponding to the n task targets, of the target materials output by the multi-task learning model;
The multi-task learning model is trained based on task information of a material sample and n task target labels of the material sample corresponding to n task targets.
In one possible implementation, the task information includes at least one of: material information, recommended user information, and recommended scene information.
In summary, after obtaining multiple predicted values of multiple materials to be recommended, which correspond to multiple task targets, the material recommendation device provided by the embodiment of the application sorts the multiple materials to be recommended according to the sorting basis of the task targets to obtain the material sequences corresponding to the task targets, normalizes the material ranks of the materials to be recommended in the material sequences, calculates the fusion score based on the multiple normalized scores of the materials to be recommended, and generates the recommendation sequence based on the fusion score sorting of the materials to be recommended, so as to recommend the materials according to the recommendation sequence; in the process, the fusion score calculation is carried out on the materials to be recommended by converting the numerical value magnitude relation between the predicted values corresponding to the multitask targets into the front-back relation of the material ranking in the material sequence corresponding to the multitask targets, so that the evaluation of the multitask targets is in the same dimension, the problems caused by different meanings and larger dimension and distribution differences among different task targets are avoided, the accuracy of material scoring is improved, and the material recommending effect is further improved.
Fig. 5 shows a block diagram of a computer device 500 according to an exemplary embodiment of the application. The computer apparatus 500 includes a central processing unit (Central Processing Unit, CPU) 501, a system memory 504 including a random access memory (Random Access Memory, RAM) 502 and a read-only memory (ROM) 503, and a system bus 505 connecting the system memory 504 and the central processing unit 501. The computer device 500 also includes a mass storage device 506 for storing an operating system 509, application programs 510, and other program modules 511.
The computer readable medium may include computer storage media and communication media without loss of generality. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, erasable programmable read-only register (Erasable Programmable Read Only Memory, EPROM), electrically erasable programmable read-only memory (EEPROM) flash memory or other solid state memory technology, CD-ROM, digital versatile disks (DIGITAL VERSATILE DISC, DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage medium is not limited to the one described above. The system memory 504 and mass storage 506 described above may be collectively referred to as memory.
According to various embodiments of the application, the computer device 500 may also operate by being connected to a remote computer on a network, such as the Internet. I.e. the computer device 500 may be connected to a network 508 via a network interface unit 507 connected to the system bus 505, or alternatively, the network interface unit 507 may be used to connect to other types of networks or remote computer systems (not shown).
The memory further includes at least one instruction, at least one program, a code set, or an instruction set, where the at least one instruction, the at least one program, the code set, or the instruction set is stored in the memory, and the central processor 501 implements all or part of the steps in the material recommendation method shown in the foregoing embodiments by executing the at least one instruction, the at least one program, the code set, or the instruction set.
Fig. 6 illustrates a block diagram of a computer device 600 in accordance with an exemplary embodiment of the present application.
In general, the computer device 600 includes: a processor 601 and a memory 602.
In some embodiments, the computer device 600 may further optionally include: a peripheral interface 603, and at least one peripheral. The processor 601, memory 602, and peripheral interface 603 may be connected by a bus or signal line. The individual peripheral devices may be connected to the peripheral device interface 603 via buses, signal lines or a circuit board. Specifically, the peripheral device includes: at least one of radio frequency circuitry 604, a display 605, a camera assembly 606, audio circuitry 607, and a power supply 608.
In some embodiments, the computer device 600 further includes one or more sensors 609. The one or more sensors 609 include, but are not limited to: acceleration sensor 610, gyroscope sensor 611, pressure sensor 612, optical sensor 613, and proximity sensor 614.
Those skilled in the art will appreciate that the architecture shown in fig. 6 is not limiting as to the computer device 600, and may include more or fewer components than shown, or may combine certain components, or employ a different arrangement of components.
In an exemplary embodiment, a computer readable storage medium is also provided, in which at least one computer program is stored, which is loaded and executed by a processor to implement all or part of the steps in the above-described material recommendation method. For example, the computer readable storage medium may be read-only memory (ROM), random-access memory (Random Access Memory, RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, optical data storage device, and the like.
In an exemplary embodiment, a computer program product is also provided, which comprises at least one computer program loaded by a processor and performing all or part of the steps of the material recommendation method as described in any of the embodiments of fig. 1 or 2 above.
Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
It is to be understood that the application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the application is limited only by the appended claims.
Claims (10)
1. A method for recommending materials, the method comprising:
Obtaining n predicted values of m materials to be recommended, wherein the m predicted values are respectively corresponding to n task targets, m is more than or equal to 2, n is more than or equal to 2, and m and n are positive integers;
respectively sequencing the m materials to be recommended based on the corresponding predicted values by taking each task target as a sequencing basis to obtain n material sequences corresponding to the n task targets;
Calculating fusion scores corresponding to the m materials to be recommended respectively based on the material ranks of the m materials to be recommended in the n material sequences respectively;
generating recommendation sequences of the m materials to be recommended based on the fusion scores corresponding to the m materials to be recommended respectively;
and recommending materials based on the recommendation sequence.
2. The method of claim 1, wherein the calculating the fusion score of each of the m materials to be recommended based on the material ranks of the m materials to be recommended in the n material sequences, respectively, comprises:
Respectively carrying out normalization processing on the material ranks in the n material sequences by taking the material sequences as units to obtain n normalization scores of the m materials to be recommended, wherein the n normalization scores correspond to n task targets respectively;
and calculating the fusion scores corresponding to the m materials to be recommended based on n normalized scores of the m materials to be recommended.
3. The method of claim 2, wherein the calculating the fused score for each of the m materials to be recommended based on n normalized scores for each of the m materials to be recommended comprises:
Carrying out score fusion on n normalized scores of a target material based on a target operation rule to obtain fusion scores of the target material; the target material is any one of the m materials to be recommended; the target operation rule is used for indicating a score fusion mode and a weighted weight value of a normalized score corresponding to each task target.
4. The method according to claim 2, wherein the normalizing the ranking of the materials in the n material sequences by taking the material sequence as a unit, to obtain n normalized scores of the m materials to be recommended, where the n normalized scores correspond to n task targets, respectively, includes:
Acquiring material ranks corresponding to the m materials to be recommended in a target material sequence; the target material sequence is any one of the n material sequences;
Carrying out normalization calculation on the material ranking in the target material sequence based on a normalization formula to obtain normalization scores of the m materials to be recommended in the target material sequence;
The higher the ranking of the materials to be recommended in the material sequence is, the higher the corresponding normalized score is.
5. The method of claim 1, wherein the recommending materials based on the recommendation sequence comprises:
Under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the large to the small of the corresponding predicted values, recommending the first K materials to be recommended with the highest fusion score in the recommended sequence;
and recommending the first K materials to be recommended with the lowest fusion score in the recommendation sequence under the condition that the n material sequences are obtained by sequencing the m materials to be recommended according to the sequence from the small to the large of the corresponding predicted values.
6. The method of claim 1, wherein the obtaining n predicted values of the m materials to be recommended, each corresponding to n task goals, comprises:
Task information of a target material is obtained; the target material is any one of the m materials to be recommended;
Inputting task information of the target materials into a multi-task learning model to obtain n predicted values, corresponding to the n task targets, of the target materials output by the multi-task learning model;
The multi-task learning model is trained based on task information of a material sample and n task target labels of the material sample corresponding to n task targets.
7. The method of claim 6, wherein the task information includes at least one of: material information, recommended user information, and recommended scene information.
8. A material recommendation device, the device comprising:
The acquisition module is used for acquiring n predicted values of m materials to be recommended, wherein the m predicted values are respectively corresponding to n task targets, m is more than or equal to 2, n is more than or equal to 2, and m and n are positive integers;
The sorting module is used for sorting the m materials to be recommended based on the corresponding predicted values by taking the task targets as sorting basis to obtain n material sequences corresponding to the n task targets;
The score calculation module is used for calculating fusion scores corresponding to the m materials to be recommended respectively based on the material ranks of the m materials to be recommended in the n material sequences respectively;
the sequence generation module is used for generating recommendation sequences of the m materials to be recommended based on the fusion scores corresponding to the m materials to be recommended respectively;
and the material recommending module is used for recommending materials based on the recommending sequence.
9. A computer device, characterized in that it comprises a processor and a memory, said memory storing at least one computer program, said at least one computer program being loaded and executed by said processor to implement the material recommendation method according to any of claims 1 to 7.
10. A computer readable storage medium, wherein at least one computer program is stored in the computer readable storage medium, the computer program being loaded and executed by a processor to implement the material recommendation method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311810390.4A CN117893278A (en) | 2023-12-26 | 2023-12-26 | Material recommendation method and device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311810390.4A CN117893278A (en) | 2023-12-26 | 2023-12-26 | Material recommendation method and device, computer equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117893278A true CN117893278A (en) | 2024-04-16 |
Family
ID=90648124
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311810390.4A Pending CN117893278A (en) | 2023-12-26 | 2023-12-26 | Material recommendation method and device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117893278A (en) |
-
2023
- 2023-12-26 CN CN202311810390.4A patent/CN117893278A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110929052B (en) | Multimedia resource recommendation method and device, electronic equipment and storage medium | |
| CN109902708B (en) | Recommendation model training method and related device | |
| CN113508378A (en) | Recommendation model training method, recommendation device and computer readable medium | |
| US20230036644A1 (en) | Method and system for exploring a personal interest space | |
| CN110796477A (en) | Advertisement display method and device, electronic equipment and readable storage medium | |
| CN111125574A (en) | Method and apparatus for generating information | |
| US8560490B2 (en) | Collaborative networking with optimized inter-domain information quality assessment | |
| CN110766513A (en) | Information sorting method and device, electronic equipment and readable storage medium | |
| CN110766510A (en) | Recommendation method and device, electronic equipment and readable storage medium | |
| CN111783810A (en) | Method and apparatus for determining attribute information of user | |
| WO2022156589A1 (en) | Method and device for determining live broadcast click rate | |
| CN116910373B (en) | House source recommendation method and device, electronic equipment and storage medium | |
| CN113204699B (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN109460778B (en) | Activity evaluation method, activity evaluation device, electronic equipment and storage medium | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| CN117893278A (en) | Material recommendation method and device, computer equipment and storage medium | |
| Kliangkhlao et al. | The design and development of a causal Bayesian networks model for the explanation of agricultural supply chains | |
| CN111368195A (en) | Model training method, device, equipment and storage medium | |
| CN110717653A (en) | Risk identification method and device and electronic equipment | |
| CN116739665A (en) | Information delivery method and device, electronic equipment and storage medium | |
| Krause et al. | Beyond the rating matrix: debiasing implicit feedback loops in collaborative filtering | |
| CN111784091A (en) | Method and apparatus for processing information | |
| CN112435091B (en) | Recommended content selection method, device, equipment and storage medium | |
| CN115907272B (en) | Method and device for evaluating brokers, electronic equipment and storage medium | |
| CN113779414B (en) | Data recommendation method, device, equipment and medium based on machine learning model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |