Multi-model ensemble method and device based on collaborative contribution rate, equipment and medium

Abstract

The application provides a multi-model integration method and device based on a synergistic contribution rate, equipment and a medium. The method comprises the following steps: obtaining a verification sample set comprising a plurality of verification samples and a multi-model set comprising a plurality of base models, and generating a plurality of first classification results by traversing the verification sample set; obtaining a plurality of comparison results based on the true label of any verification sample and all the first classification results, and assigning a contribution rate value to each base model based on all the comparison results; generating a synergistic contribution rate based on any base model and all the contribution rate values corresponding to the base model; obtaining a positive score and a negative score of a target model based on a target sample and all the synergistic contribution rates, and determining the target sample based on the positive score and the negative score. According to the technical scheme of the embodiment of the application, the fusion weights of the base models can be quantified by the synergistic contribution rate, so that the overall performance and efficiency of the multi-model integration are improved.

Description

Technical Field

[0001] This invention relates to the field of multi-model integration technology, and in particular to a multi-model integration method, apparatus, device, and medium based on collaborative contribution rate. Background Technology

[0002] In the process of repeatedly classifying similar samples, pre-trained models are usually used to classify the samples to avoid repetitive work. However, due to limitations such as insufficient generalization ability of a single model, the classification results output by a single model are biased and have low accuracy.

[0003] Therefore, existing technologies often employ Bayesian ensemble learning to fuse multiple models, thereby achieving a multi-model ensemble. Traditional Bayesian ensemble learning is based on the Naive Bayes assumption that models are independent. After fusion, there are multiple base models (base classifiers), each of which obtains a binary classification result (positive or negative) from the samples. Based on all the binary classification results, positive and negative scores are calculated separately. When the positive score is large, the multi-model fusion outputs a positive classification result; when the negative score is large, the multi-model fusion outputs a negative score result.

[0004] However, traditional Bayesian ensemble learning methods, in the process of multi-model fusion, typically calculate probability parameters based solely on the recognition accuracy of a single model, ignoring the collaborative relationships between models. In multi-model ensemble scenarios, a large number of homogeneous models exist: some models have highly similar discrimination logic, correctly identifying most samples simultaneously; others possess independent discrimination capabilities, able to differentiate and correct erroneous samples. Traditional Bayesian algorithms cannot distinguish between homogeneous models, leading to the continuous accumulation of weights on these models, diluting the contributions of independent, high-quality models, and causing problems such as degraded ensemble performance, limited fusion accuracy, and inefficient model utilization. Summary of the Invention

[0005] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a multi-model integration method, apparatus, device, and medium based on collaborative contribution rate, which can quantify the fusion weights of each base model through the differential collaborative contribution rate, thereby improving the overall performance and efficiency of multi-model integration.

[0006] In a first aspect, embodiments of the present invention provide a multi-model ensemble method based on collaborative contribution rate, including: Obtain a validation sample set and a multi-model set. The multi-model set iterates through the validation sample set to generate multiple first classification results. The multi-model set includes multiple base models, and the validation sample set includes multiple validation samples. The first classification result is the classification result of the base models based on the validation samples. Based on any one of the verification samples, the true label of the verification sample is obtained, and multiple comparison results are obtained based on the true label and all the first classification results. A contribution rate value is assigned to each base model based on all the comparison results, wherein the true label is used to characterize the true classification result of the verification sample, and all the contribution rate values ​​corresponding to the verification sample are one. Based on any one of the base models, a collaborative contribution rate is generated based on all the contribution rate values ​​corresponding to the base model. Obtain a target sample. The multi-model set obtains the positive and negative scores of the target model based on the target sample and all the collaborative contribution rates. When the positive score is greater than the negative score, the target sample is a positive sample. Alternatively, when the negative score is greater than or equal to the positive score, the target sample is a negative sample.

[0007] According to some embodiments of the present invention, multiple comparison results are obtained based on the true labels and all the first classification results, and a contribution rate value is assigned to each of the base models based on all the comparison results, including: Obtain the true label of the verification sample and all the first classification results corresponding to the verification sample, and compare the true label with all the first classification results to obtain multiple comparison results; The base models corresponding to the same comparison results are marked as first models, and the base models corresponding to different comparison results are marked as second models. The same contribution rate value is assigned to all first models, and a zero contribution rate value is assigned to all second models, wherein the contribution rate value assigned to the first models is not zero.

[0008] According to some embodiments of the present invention, assigning the same contribution rate value to all the first models includes: Obtain a first quantity, wherein the first quantity is the total number of all the first models; A contribution rate value is assigned to all of the first models, wherein the contribution rate value of the first model is equal to one-n, and n is the first quantity.

[0009] According to some embodiments of the present invention, generating a collaborative contribution rate based on all the contribution rate values ​​corresponding to any one of the base models includes: Based on any one of the base models, obtain all the contribution rate values ​​corresponding to the base model, and sum all the contribution rate values ​​to obtain the candidate contribution rate value. The candidate contribution rate values ​​corresponding to all base models are normalized to obtain the collaborative contribution rate corresponding to all base models.

[0010] According to some embodiments of the present invention, the multi-model set traverses the validation sample set to generate multiple first classification results, including: All base models are iterated through all the verification samples; Based on any one of the base models, the base model outputs multiple first classification results based on all the validation samples.

[0011] According to some embodiments of the present invention, the multi-model ensemble obtains the positive and negative scores of the target model based on the target samples and all the collaborative contribution rates, including: Based on any one of the base models, the base model obtains the target sample. When the base model determines that the target sample is a positive sample, the first parameter of the base model is equal to the positive class correct probability term multiplied by the collaborative contribution rate, and the second parameter of the base model is equal to the negative class error probability term. Alternatively, when the base model determines that the target sample is a negative sample, the third parameter of the base model is equal to the negative class correct probability term multiplied by the collaborative contribution rate, and the fourth parameter of the base model is equal to the positive class error probability term. The positive score and the negative score are obtained based on the first parameter, the second parameter, the third parameter, and the fourth parameter.

[0012] In a second aspect, embodiments of the present invention provide a multi-model integration apparatus based on collaborative contribution rate, including at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform the multi-model integration method based on collaborative contribution rate as described in the first aspect above.

[0013] Thirdly, embodiments of the present invention provide an electronic device including a multi-model integration device based on collaborative contribution rate as described in the second aspect above.

[0014] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions for executing the multi-model integration method based on collaborative contribution rate as described in the first aspect above.

[0015] The multi-model ensemble method based on collaborative contribution rate according to embodiments of the present invention has at least the following beneficial effects: A validation sample set and a multi-model set are obtained, wherein the multi-model set traverses the validation sample set to generate multiple first classification results, wherein the multi-model set includes multiple base models, the validation sample set includes multiple validation samples, and the first classification results are the classification results of the base models based on the validation samples; Based on any one of the validation samples, the true label of the validation sample is obtained, multiple comparison results are obtained based on the true label and all the first classification results, and a contribution rate value is assigned to each base model based on all the comparison results, wherein the true label is used to characterize the true classification result of the validation sample, and all the contribution rate values ​​corresponding to the validation sample are one; Based on any one of the base models, a collaborative contribution rate is generated based on all the contribution rate values ​​corresponding to the base model; A target sample is obtained, wherein the multi-model set obtains the positive score and negative score of the target model based on the target sample and all the collaborative contribution rates, wherein when the positive score is greater than the negative score, the target sample is a positive sample, or when the negative score is greater than or equal to the positive score, the target sample is a negative sample. According to the technical solution of the present invention, multiple comparison results are output through all first classification results and corresponding real labels, and a contribution rate value is assigned to each base model. In order to reduce the influence of homogeneous base models and increase the weight of base models with independent discrimination capabilities, and the sum of the contribution rate values ​​corresponding to each validation sample is one, that is, when homogeneous base models make the same correct judgment, their contribution rate values ​​will also be reduced by the spread. However, when a base model with independent discrimination capabilities makes a correct judgment and a homogeneous base model makes an incorrect judgment, only the base model with independent discrimination capabilities will spread the contribution rate value. Thus, after traversing the validation sample set, the total contribution rate value obtained by the base model with independent discrimination capabilities is higher, that is, its collaborative contribution rate is higher, thereby amplifying its proportion in the multi-model set, weakening the interference of homogeneous and redundant base models, and thus improving the overall performance and efficiency of multi-model integration. Attached Figure Description

[0016] Figure 1 This is a flowchart of a multi-model ensemble method based on collaborative contribution rate provided in an embodiment of the present invention; Figure 2 This is a structural diagram of a multi-model integration device based on collaborative contribution rate provided in another embodiment of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0019] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0020] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0021] This invention provides a multi-model ensemble method, apparatus, device, and medium based on collaborative contribution rate. The multi-model ensemble method includes: acquiring a validation sample set and a multi-model set; the multi-model set iterating through the validation sample set to generate multiple first classification results, wherein the multi-model set includes multiple base models, the validation sample set includes multiple validation samples, and the first classification results are the classification results of the base models based on the validation samples; acquiring the true label of any one of the validation samples; obtaining multiple comparison results based on the true label and all the first classification results; assigning a contribution rate value to each base model based on all the comparison results, wherein the true label is used to characterize the true classification result of the validation sample, and all the contribution rate values ​​corresponding to the validation sample are one; generating a collaborative contribution rate based on any one of the base models and all the contribution rate values ​​corresponding to the base model; acquiring a target sample; the multi-model set obtaining a positive score and a negative score of the target model based on the target sample and all the collaborative contribution rates; when the positive score is greater than the negative score, the target sample is a positive sample, or when the negative score is greater than or equal to the positive score, the target sample is a negative sample. According to the technical solution of the present invention, multiple comparison results are output through all first classification results and corresponding real labels, and a contribution rate value is assigned to each base model. In order to reduce the influence of homogeneous base models and increase the weight of base models with independent discrimination capabilities, and the sum of the contribution rate values ​​corresponding to each validation sample is one, that is, when homogeneous base models make the same correct judgment, their contribution rate values ​​will also be reduced by the spread. However, when a base model with independent discrimination capabilities makes a correct judgment and a homogeneous base model makes an incorrect judgment, only the base model with independent discrimination capabilities will spread the contribution rate value. Thus, after traversing the validation sample set, the total contribution rate value obtained by the base model with independent discrimination capabilities is higher, that is, its collaborative contribution rate is higher, thereby amplifying its proportion in the multi-model set, weakening the interference of homogeneous and redundant base models, and thus improving the overall performance and efficiency of multi-model integration.

[0022] The technical solutions of the embodiments of the present invention will be further illustrated in the following figures.

[0023] Reference Figure 1 , Figure 1 A flowchart of a multi-model ensemble method based on collaborative contribution rate provided by an embodiment of the present invention, the multi-model ensemble method based on collaborative contribution rate includes, but is not limited to, the following steps: S10: Obtain the validation sample set and the multi-model set. The multi-model set iterates through the validation sample set to generate multiple first classification results. The multi-model set includes multiple base models, and the validation sample set includes multiple validation samples. The first classification result is the classification result of the base models based on the validation samples.

[0024] It should be noted that the multiple model set traverses the validation sample set, meaning all base models traverse all validation samples to generate multiple first classification results. The base models classify the validation samples into positive and negative samples, and the first classification results include both positive and negative samples. For any given validation sample, the validation sample passes through all base models, and the first classification result output by the base models is recorded.

[0025] S20: Based on any validation sample, obtain the true label of the validation sample, obtain multiple comparison results based on the true label and all first classification results, and assign a contribution rate value to each base model based on all comparison results. Here, the true label is used to characterize the true classification result of the validation sample, and the total contribution rate value corresponding to the validation sample is one.

[0026] It should be noted that, for any validation sample, the base model whose first classification result output by the base model is consistent with the true label of the validation sample is the model whose first classification result is correct.

[0027] It should be noted that the first classification results output by each base model based on the same validation samples are not necessarily completely the same. The true labels representing the true classification results of the validation samples are compared with all the first classification results individually to obtain multiple comparison results. When the first classification result is the same as the true label, the comparison result is the same; when the first classification result is different from the true label, the comparison result is different. The comparison results include the same and different.

[0028] In existing technologies, multi-model ensembles also include multiple base models. Traditional Bayesian ensemble learning, based on the Naive Bayes assumption of model independence, typically calculates probability parameters based solely on the recognition accuracy of a single base model during multi-model fusion. In multi-model ensemble scenarios based on traditional Bayesian ensemble learning, a large number of homogeneous base models exist: some base models have highly similar discrimination logic, and most validation samples are correctly judged simultaneously; some models possess independent discrimination capabilities, able to differentiate and correct erroneous samples. However, existing technologies cannot prove the independence of each base model or quantify its correlation, ignoring the collaborative relationships between base models. Traditional Bayesian algorithms cannot distinguish between the redundancy and independent contribution of base models. The continuous accumulation of weights on homogeneous base models dilutes the contribution of independent, high-quality models, easily leading to problems such as degraded ensemble performance, limited fusion accuracy, and unreasonable model utilization.

[0029] It should be noted that in this application, the comparison results obtained through the first classification result and the true label are used to determine whether the judgment of each base model is correct or incorrect, and a contribution rate value is assigned to the base model based on the comparison results. Homogeneous base models, based on the same validation samples, output first classification results and comparison results that are essentially the same. When the first classification results output by the homogeneous base model and the independent high-quality model are different, a higher contribution rate value is assigned to the model with the same comparison result (i.e., the independent high-quality model), and a lower or zero contribution rate value is assigned to the model with different comparison results (i.e., the homogeneous model). This results in the independent high-quality model receiving a higher contribution rate value, improving fusion accuracy, and enhancing the overall performance and efficiency of multi-model integration.

[0030] S30 generates a collaborative contribution rate based on any base model and all contribution rate values ​​corresponding to the base model.

[0031] It should be noted that because homogeneous models often make simultaneous judgments, their contribution rate is lower when the corresponding comparison results are the same. Independent high-quality models, with their independent discrimination capabilities, rely less on the group and have stronger independent error correction capabilities. When the comparison results of independent high-quality models are the same while those of homogeneous models are different, the contribution rate of independent high-quality models is higher. After traversing a validation sample set including multiple validation samples, the contribution rate obtained by independent high-quality models will inevitably be higher, meaning their collaborative contribution rate is higher.

[0032] S40, Obtain the target sample. The multi-model set obtains the positive and negative scores of the target model based on the target sample and the total collaborative contribution rate. When the positive score is greater than the negative score, the target sample is a positive sample; or, when the negative score is greater than or equal to the positive score, the target sample is a negative sample.

[0033] It should be noted that the expression for the positive score formula in the existing formula is: The expression for negative scores is: Based on traditional Bayesian evidence accumulation logic, this paper introduces a collaborative contribution rate to reconstruct the positive and negative score calculation rules, distinguishing between the calculation logic of correct and incorrect terms in the model. This allows for the generation of positive and negative scores for each target sample based on all target samples and their corresponding collaborative contribution rates. However, existing techniques do not address how to determine whether a target sample is positive or negative based solely on its positive and negative scores.

[0034] It should be noted that weighted enhancement based on the collaborative contribution rate is only applied when the base model's judgment and corresponding score are the same. When the base model's judgment and corresponding score are different, the original coefficients remain unchanged. That is, when the base model outputs a positive sample judgment, the positive class correct probability term is multiplied by the collaborative contribution rate for weighted enhancement, while the negative class incorrect probability term retains its original coefficient; when the base model outputs a negative sample judgment, the negative class correct probability term is multiplied by the collaborative contribution rate. i Weighted enhancement is applied while keeping the original coefficients of the positive class error probability term unchanged. Through the differentiated weighting method of this application, the evidence proportion of independent, high-quality models is amplified, while the interference from homogeneous, redundant models is weakened, thus achieving multi-model fusion inference.

[0035] In existing technologies, traditional Bayesian ensemble learning relies on the Naive Bayes assumption of model independence. During multi-model fusion, traditional methods typically calculate probability parameters based solely on the individual model's accuracy, neglecting the collaborative relationships between models. In multi-model ensemble scenarios, many models are homogeneous, some possessing independent discriminative capabilities and able to differentiate and correct erroneous samples. For example, some models exhibit highly similar discriminative logic, correctly identifying most samples simultaneously. However, in reality, the independence of each base model cannot be proven, nor can the correlation be quantified. Traditional Bayesian algorithms cannot distinguish between model redundancy and independent contribution. The continuous accumulation of weights on homogeneous models dilutes the contribution of independent, high-quality models, easily leading to degraded ensemble performance, limited fusion accuracy, and unreasonable model utilization. The core parameters of traditional Bayesian ensemble learning include Ri+, Ei+, and Ri... and Ei Ri+ represents the probability that model i outputs the correct value when the true label is positive; Ei+ represents the probability that model i outputs the incorrect value when the true label is positive; Ri When the true label is negative, model i outputs the probability that the label is correct; Ei Let $\mathbf{i}$ be the probability that model $i$ outputs an incorrect value when the true label is negative. The formula for positive score is: The expression for negative scores is: In existing multi-model ensembles, each base model is a base classifier, and each base classifier outputs a binary classification result (positive or negative) based on the sample; for each sample, a positive score (i.e., (P(+|all model outputs))) and a negative score (i.e., (P(...))) are calculated. |All model outputs))) When the positive score is greater than the negative score, the sample is determined to be a positive sample; or, when the negative score is greater than the positive score, the sample is determined to be a negative sample.

[0036] In this application, the traditional Bayesian ensemble logic is optimized by incorporating a collaborative contribution rate calculation mechanism, achieving differentiated quantification of the fusion weights of each model. This method boasts strong algorithmic versatility, independent of specific base model structures; it automatically suppresses homogeneous redundant models, amplifies the contribution of independently discriminant models, and addresses the degradation problem of multi-model ensemble performance; the contribution rate is entirely generated through statistical data, without manual hyperparameters, ensuring objective and robust weight allocation; it retains the interpretability of traditional Bayesian probabilities while optimizing the fusion logic, stably improving ensemble classification accuracy without increasing inference complexity. It effectively addresses the homogeneity and redundancy problem of base models through the collaborative contribution rate, quantifies the relevance of base models, and integrates it into the Bayesian probability calculation of ensemble learning, thereby improving overall performance and efficiency.

[0037] Additionally, in one embodiment, in Figure 1 In step S20, multiple comparison results are obtained based on the true labels and all first classification results. A contribution rate value is assigned to each base model based on all comparison results, including but not limited to the following steps: S21, obtain the true label of the verification sample and all the first classification results corresponding to the verification sample, and compare the true label with all the first classification results to obtain multiple comparison results; S22, mark the base models with the same comparison results as the first model, mark the base models with different comparison results as the second model, assign the same contribution rate value to all first models, and assign a contribution rate value of zero to all second models, wherein the contribution rate value assigned to the first model is not zero.

[0038] It should be noted that the method of this application has a model collaborative contribution rate calculation mechanism, which obtains multiple comparison results based on all base models and the validation sample set, thereby statistically analyzing the collaborative discriminant relationship between each base model and calculating the collaborative contribution rate of each base model.

[0039] It should be noted that the process involves iterating through all validation samples, recording the first classification result (output result) and the true label of all base models; for a single validation sample, selecting base models whose output comparison results are the same / correct, and counting the number of currently correct models, n; for all models correctly judged in this sample, the contribution rate value 1 / n is accumulated. After completing the statistics for all samples, the collaborative contribution rate of each base model is obtained by normalization.

[0040] It should be noted that homogeneous models are often judged to be correct simultaneously. Since contribution rate values ​​are only assigned to base models with the same comparison results, the contribution rate value allocated to homogeneous models is relatively low. Models with independent discrimination capabilities rely less on the group and have strong independent error correction capabilities. They can output the correct comparison result when homogeneous models make mistakes. In other words, the number of first models is smaller, which means there are fewer base models to allocate contribution rate values, thus enabling them to obtain a higher contribution rate.

[0041] In another embodiment, in step S22, all first models are assigned the same contribution rate value, including but not limited to the following steps: S221, obtain the first quantity, where the first quantity is the total number of all first models; S222, assign contribution rate values ​​to all first models, where the contribution rate value of the first model is equal to 1 / n, and n is the first quantity.

[0042] It should be noted that each first model is assigned the same contribution rate, and the sum of the contribution rate values ​​corresponding to each validation sample is one. Therefore, the contribution rate value of the first model assigned to each validation sample is 1 / n. For a single validation sample, the base models (i.e., the first models) that output the correct judgment result (i.e., the comparison result is the same) are selected, and the number of the current first models (i.e., the first number, which is n) is counted. Contribution rate values ​​are assigned to all first models corresponding to the validation sample.

[0043] It should be noted that this application proposes a mechanism for quantifying the collaborative contribution rate of base models. Unlike traditional Bayesian methods that rely solely on the accuracy of a single model to assign weights, this invention designs a method for calculating the collaborative contribution rate from the perspective of the collaborative relationship between base models. By statistically analyzing the common correct frequency of base models at the sample level, it automatically distinguishes between homogeneous base models and independently discriminative base models; homogeneous models are assigned lower weights, while independently discriminative models receive higher weights, thereby objectively quantifying the irreplaceability of each model in the integrated system.

[0044] Additionally, in one embodiment, in Figure 1 Step S30 shown includes, but is not limited to, the following steps: S31. Based on any base model, obtain all contribution rate values ​​corresponding to the base model, and sum all contribution rate values ​​to obtain candidate contribution rate values. S32, normalize the candidate contribution rate values ​​corresponding to all base models to obtain the collaborative contribution rate corresponding to all base models.

[0045] It should be noted that after all validation samples have completed the contribution rate value allocation, the contribution rate values ​​obtained for each base model are statistically analyzed to obtain candidate contribution rate values. For ease of calculation, the candidate contribution rate values ​​are normalized to obtain the collaborative contribution rate of each base model. That is, for a single base model, the higher the sum of the obtained contribution rate values, i.e., the higher the candidate contribution rate value, and thus the higher the corresponding collaborative contribution rate.

[0046] Additionally, in one embodiment, in Figure 1 In step S10 shown, the multi-model set traverses the validation sample set to generate multiple first classification results, including but not limited to the following steps: S11, traverse all validation samples of the entire base model; S12 is based on any base model, and the base model outputs multiple first classification results based on all validation samples.

[0047] It should be noted that each base model acquires all validation samples and outputs the corresponding first classification result. That is, the number of first classification results is the number of base models multiplied by the number of validation samples, and the number of first classification results is equal to the number of comparison results.

[0048] Additionally, in one embodiment, in Figure 1 In step S40 shown, the multi-model ensemble obtains the positive and negative scores of the target model based on the target samples and the total collaborative contribution rate, including but not limited to the following steps: S41, based on any base model, the base model obtains the target sample. When the base model determines that the target sample is a positive sample, the first parameter of the base model is equal to the positive class correct probability term multiplied by the collaborative contribution rate, and the second parameter of the base model is equal to the negative class error probability term. S42, or, when the base model determines that the target sample is a negative sample, the third parameter of the base model is equal to the negative class correct probability term multiplied by the collaborative contribution rate, and the fourth parameter of the base model is equal to the positive class error probability term; S43, positive and negative scores are obtained based on the first, second, third and fourth parameters.

[0049] It should be noted that the method of this application has a Bayesian fusion calculation term with contribution rate constraint. Based on the traditional Bayesian evidence accumulation logic in the prior art, it introduces a collaborative contribution rate reconstruction positive and negative score calculation rule to distinguish the calculation logic of correct and incorrect terms of the base model.

[0050] It should be noted that when the base model output determines the target sample as a positive sample, the positive class correct probability term is multiplied by the collaborative contribution rate α. i Weighted augmentation is applied to obtain positive scores, while the negative class error probability term retains its original coefficient (i.e., 1) to obtain negative scores; when the model output classifies a sample as negative, the negative class correct probability term is multiplied by the co-contribution rate α. i Weighted enhancement is applied while keeping the original coefficients of the positive class error probability term unchanged. By using differentiated weighting, the proportion of evidence from independent, high-quality models is amplified, while the interference from homogeneous, redundant models is mitigated.

[0051] Within the traditional Bayesian evidence accumulation framework, this paper innovatively distinguishes the weighting logic between correct and incorrect terms: only the probability terms that the model determines are enhanced with a collaborative contribution rate, while incorrect terms are not weighted. This avoids interference from inferior and redundant models with the fusion results, thereby correcting the defect of excessively high weights in homogeneous models in traditional Bayesian ensemble and improving the overall performance and efficiency of multi-model ensemble.

[0052] In this application, when the base model output is positive, the cumulative product of positive scores is Ri+×αi (correct terms × contribution rate), and the cumulative product of negative scores is Ei. (Error items, not multiplied by contribution rate); When the base model output is negative, the cumulative negative scores are Ri. ×αi (correct items × contribution rate), positive scores are accumulated and multiplied to Ei+ (incorrect items, without multiplying by contribution rate).

[0053] like Figure 2 As shown, Figure 2 This is a structural diagram of a multi-model ensemble device based on cooperative contribution rate provided in one embodiment of the present invention. The present invention also provides a multi-model ensemble device based on cooperative contribution rate, comprising: The processor 501 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 502 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 502 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 502 and is called and executed by the processor 501 using the multi-model integration method based on collaborative contribution rate of the embodiments of this application. The input / output interface 503 is used to implement information input and output; The communication interface 504 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 505 transmits information between various components of the device (e.g., processor 501, memory 502, input / output interface 503, and communication interface 504); The processor 501, memory 502, input / output interface 503, and communication interface 504 are connected to each other within the device via bus 505.

[0054] This application also provides an electronic device, including the multi-model integration device based on the collaborative contribution rate as described above.

[0055] This application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described multi-model integration method based on collaborative contribution rate.

[0056] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate, and may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0057] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0058] The above provides a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A multi-model ensemble method based on collaborative contribution rate, characterized in that, include: Obtain a validation sample set and a multi-model set. The multi-model set iterates through the validation sample set to generate multiple first classification results. The multi-model set includes multiple base models, and the validation sample set includes multiple validation samples. The first classification result is the classification result of the base models based on the validation samples. Based on any one of the verification samples, the true label of the verification sample is obtained, and multiple comparison results are obtained based on the true label and all the first classification results. A contribution rate value is assigned to each base model based on all the comparison results, wherein the true label is used to characterize the true classification result of the verification sample, and all the contribution rate values ​​corresponding to the verification sample are one. Based on any one of the base models, a collaborative contribution rate is generated based on all the contribution rate values ​​corresponding to the base model. Obtain a target sample. The multi-model set obtains the positive and negative scores of the target model based on the target sample and all the collaborative contribution rates. When the positive score is greater than the negative score, the target sample is a positive sample. Alternatively, when the negative score is greater than or equal to the positive score, the target sample is a negative sample.

2. The multi-model ensemble method based on collaborative contribution rate according to claim 1, characterized in that, Multiple comparison results are obtained based on the true labels and all the first classification results. A contribution rate value is assigned to each base model based on all the comparison results, including: Obtain the true label of the verification sample and all the first classification results corresponding to the verification sample, and compare the true label with all the first classification results to obtain multiple comparison results; The base models corresponding to the same comparison results are marked as first models, and the base models corresponding to different comparison results are marked as second models. The same contribution rate value is assigned to all first models, and a zero contribution rate value is assigned to all second models, wherein the contribution rate value assigned to the first models is not zero.

3. The multi-model ensemble method based on collaborative contribution rate according to claim 2, characterized in that, Assigning the same contribution rate value to all of the first models includes: Obtain a first quantity, wherein the first quantity is the total number of all the first models; A contribution rate value is assigned to all of the first models, wherein the contribution rate value of the first model is equal to one-n, and n is the first quantity.

4. The multi-model ensemble method based on collaborative contribution rate according to claim 1, characterized in that, Based on any one of the base models, a collaborative contribution rate is generated based on all the contribution rate values ​​corresponding to the base model, including: Based on any one of the base models, obtain all the contribution rate values ​​corresponding to the base model, and sum all the contribution rate values ​​to obtain the candidate contribution rate value. The candidate contribution rate values ​​corresponding to all base models are normalized to obtain the collaborative contribution rate corresponding to all base models.

5. The multi-model ensemble method based on collaborative contribution rate according to claim 1, characterized in that, The multi-model set traverses the validation sample set to generate multiple first classification results, including: All base models are iterated through all the verification samples; Based on any one of the base models, the base model outputs multiple first classification results based on all the validation samples.

6. The multi-model ensemble method based on collaborative contribution rate according to claim 1, characterized in that, The multi-model ensemble obtains the positive and negative scores of the target model based on the target samples and all the collaborative contribution rates, including: Based on any one of the base models, the base model obtains the target sample. When the base model determines that the target sample is a positive sample, the first parameter of the base model is equal to the positive class correct probability term multiplied by the collaborative contribution rate, and the second parameter of the base model is equal to the negative class error probability term. Alternatively, when the base model determines that the target sample is a negative sample, the third parameter of the base model is equal to the negative class correct probability term multiplied by the collaborative contribution rate, and the fourth parameter of the base model is equal to the positive class error probability term. The positive score and the negative score are obtained based on the first parameter, the second parameter, the third parameter, and the fourth parameter.

7. A multi-model integration device based on collaborative contribution rate, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform the multi-model integration method based on collaborative contribution rate as described in any one of claims 1 to 6.

8. An electronic device, characterized in that, Includes the multi-model integration device based on collaborative contribution rate as described in claim 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the multi-model integration method based on collaborative contribution rate as described in any one of claims 1 to 6.

Images