A data processing method, device, product and medium

Abstract

The present invention discloses a data processing method, device, product and medium, and relates to the field of computer technology. The method uses multiple prediction models to predict each comment in a comment set respectively, and determines the final prediction results of the multiple prediction models for each comment, that is, based on the inference of group reactions based on the multiple prediction models, the comments of each user are aggregated, and the accuracy of the inference of group reactions is improved; when it is detected that the results predicted by the prediction models for the current comment are the same or different, different methods are used to determine the comprehensive prediction results of the comment set, which improves the efficiency of determining the comprehensive prediction results; when inferring the group reaction, the confidence of the prediction results of the prediction model is taken into account, and the confidence of the prediction results of the prediction model is at least determined by the information entropy of the probability distribution output by the prediction model, so that the confidence of the prediction results of the obtained prediction model is more accurate, and the accuracy of the obtained comprehensive prediction results is further improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a data processing method, device, product and medium. Background Art

[0002] With the development of Internet technology, various social media have gradually become the mainstream tools for people's social Internet life. The huge social media user group expresses their opinions and emotions by creating, disseminating content and posting comments on social media platforms. It is of great significance to analyze the overall reaction of a group of users to an event on social media.

[0003] The reaction of a group is the integration of the opinions and emotions of multiple single users, so a reasonable method is needed to infer the distribution of group opinions and emotions. In the inference technology of the overall reaction of related group users, a single model is often used, which often leads to biased prediction results.

[0004] It can be seen that how to more accurately infer the response of the cluster is a technical problem that people in this field urgently need to solve. Summary of the invention

[0005] The purpose of the present invention is to provide a data processing method, device, product and medium to solve the technical problem that when inferring the overall response of a group of users, the use of a single model leads to low accuracy of the prediction result.

[0006] In order to solve the above technical problems, the present invention provides a data processing method, comprising:

[0007] Obtaining a review set and inputting at least one review in the review set into a plurality of prediction models;

[0008] Get the prediction results of each prediction model for the same comment respectively;

[0009] When it is detected that the prediction results of the current comment by each prediction model are the same, the average of the prediction results of the current comment by all prediction models is obtained to determine the final prediction results of the current comment by multiple prediction models; the comprehensive prediction result of the comment set is determined according to the sum of the final prediction results of each current comment input into the prediction model;

[0010] When it is detected that the prediction results of each prediction model for the current comment are different, the final prediction results of multiple prediction models for the current comment are determined according to the prediction result of the target prediction model for the current comment and the confidence of the target prediction model prediction result, and the step of determining the comprehensive prediction result of the comment set according to the sum of the final prediction results of each current comment input into the prediction model is entered; wherein the target prediction model is a model determined from all prediction models; and the confidence of the prediction result of the prediction model is determined by at least the information entropy of the probability distribution output by the prediction model.

[0011] On the one hand, the confidence of the prediction model's prediction results is determined by the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model's prediction results;

[0012] The confidence level of the prediction model's prediction results is determined based on the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model's prediction results, including:

[0013] Get the logarithmic value corresponding to the number of classifications of the prediction results of the prediction model;

[0014] Obtain the ratio of the information entropy of the probability distribution output by the prediction model to the logarithmic value corresponding to the number of classifications of the prediction results of the prediction model;

[0015] Obtain a difference obtained by subtracting the ratio from 1, and use the difference as the confidence of the prediction result of the prediction model.

[0016] On the other hand, the categories that determine the prediction results of the prediction model include:

[0017] Determining in advance the reaction of the first user to the commented object based on the content of at least one first user's comment; wherein the reaction of the first user to the commented object includes at least an emotional reaction and an opinion reaction;

[0018] The category of the emotional response of the first user in the response to the commented object or the category of the opinion response of the first user in the response to the commented object is used as the category of the prediction result of the prediction model.

[0019] On the other hand, the obtaining of the prediction results of each prediction model for the same comment includes:

[0020] Obtain the categories predicted by each prediction model for the same comment and the probability distribution corresponding to the categories;

[0021] Determining whether the prediction results of each prediction model for the current comment are the same includes:

[0022] When it is detected that the categories predicted by the prediction models for the same comment are the same, determining that the prediction results of the prediction models for the current comment are the same;

[0023] When it is detected that the categories predicted by the prediction models for the same comment are different, and at least one group of prediction models predicts the same category for the same comment, it is determined that the prediction results of the prediction models for the current comment are not completely the same;

[0024] When it is detected that the categories predicted by the prediction models for the same comment are completely different, it is determined that the results predicted by the prediction models for the current comment are completely different.

[0025] On the other hand, when it is detected that the prediction results of the prediction models for the current comment are different, the final prediction results of the multiple prediction models for the current comment are determined according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model, including:

[0026] When it is detected that the prediction results of the prediction models for the current comment are different, the total number of prediction models is obtained;

[0027] The target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models, or the target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models;

[0028] The final prediction results of multiple prediction models for the current comment are determined according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model.

[0029] On the other hand, determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models includes:

[0030] When it is detected that the number of prediction models that predict the same result for the current comment among all the prediction models is greater than half of the total number of prediction models, the prediction model that predicts the same result for the current comment is used as the initial target prediction model;

[0031] After determining the initial target prediction model, obtaining a first confidence level of the prediction results of each initial target prediction model and a second confidence level of the prediction results of the remaining prediction models; wherein the remaining prediction models are prediction models other than the initial target prediction model among all prediction models;

[0032] Determining, according to the relationship between the second confidence level and the first confidence level, whether each of the remaining prediction models is to be used as a model in the target prediction model or not to be used as a model in the target prediction model;

[0033] If it is determined that the remaining prediction model is a model in the target prediction model, the initial target prediction model and the remaining prediction model that satisfies the relationship between the second confidence level and the first confidence level are used as the final target prediction model.

[0034] On the other hand, the determining, according to the relationship between the second confidence level and the first confidence level, whether each of the remaining prediction models is used as a model in the target prediction model or not as a model in the target prediction model includes:

[0035] Obtaining the maximum first confidence degree among all the first confidence degrees;

[0036] In the case where it is detected that the current second confidence is greater than the maximum first confidence, determining the remaining prediction model corresponding to the current second confidence as a model in the target prediction model;

[0037] When it is detected that the current second confidence is less than or equal to the maximum first confidence, it is determined that the remaining prediction model corresponding to the current second confidence is not used as a model in the target prediction model.

[0038] On the other hand, determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models includes:

[0039] When it is detected that the number of prediction models that predict the same result for the current comment among all the prediction models is less than or equal to half of the total number of prediction models, each prediction model is used as the final target prediction model.

[0040] On the other hand, before determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models, the method further includes:

[0041] Determine whether the total number of prediction models meets preset requirements; wherein the preset requirements at least include that the total number of prediction models is greater than or equal to a preset value, or that the total number of prediction models is an odd number;

[0042] If yes, then enter the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models;

[0043] If not, the prediction model is adjusted and the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models is entered, or the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models is entered.

[0044] On the other hand, the preset requirement is that the total number of prediction models is an odd number; and adjusting the prediction model includes:

[0045] Obtain the prediction model with the lowest confidence in the prediction result among all the prediction models; eliminate the prediction model with the lowest confidence in the prediction result;

[0046] Or, obtain an odd number of new prediction models, and add the new prediction models to the existing prediction models.

[0047] On the other hand, the preset requirement is that the total number of prediction models is greater than or equal to a preset value; and adjusting the prediction model includes:

[0048] When the total number of detected prediction models is less than the preset value, obtaining a new prediction model;

[0049] Add new prediction models to existing prediction models.

[0050] On the other hand, determining the comprehensive prediction result of the review set according to the sum of the final prediction results of each current review input into the prediction model includes:

[0051] Obtain the second user's response to the first user's current comment;

[0052] Determine the weight of the current comment according to the reaction of the second user to the current comment of the first user;

[0053] Obtain the product of the final prediction result of each current comment and the weight of the current comment, and use the product as the new final prediction result of the current comment;

[0054] The comprehensive prediction result of the review set is determined according to the sum of the new final prediction results of each current review input into the prediction model.

[0055] In order to solve the above technical problems, the present invention further provides a data processing device, comprising:

[0056] Memory for storing computer programs;

[0057] A processor is used to implement the steps of the above-mentioned data processing method when executing the computer program.

[0058] In order to solve the above technical problem, the present invention also provides a computer program product, including a computer program / instruction, which implements the steps of the above data processing method when executed by a processor.

[0059] In order to solve the above technical problem, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above data processing method are implemented.

[0060] The data processing method provided by the present invention includes: obtaining a review set and inputting at least one review in the review set into multiple prediction models; respectively obtaining the results predicted by each prediction model for the same review; and when it is detected that the results predicted by each prediction model for the current review are the same or different, using different methods to determine the comprehensive prediction results of the review set.

[0061] The beneficial effects of the present invention are as follows: firstly, compared with the method of using a single model to infer group reactions, the method provided by the present invention uses multiple prediction models to predict each comment in the comment set respectively, and determines the final prediction results of the multiple prediction models for each comment, that is, the group reaction is inferred according to the multiple prediction models, thereby realizing the aggregation of the comments of each user and improving the accuracy of the group reaction inference; secondly, when it is detected that the results predicted by the prediction models for the current comment are the same or different, different methods are used to determine the comprehensive prediction results of the comment set, thereby improving the efficiency of determining the comprehensive prediction results, such as when it is detected that the prediction models for the current comment are the same or different, different methods are used to determine the comprehensive prediction results of the comment set, thereby improving the efficiency of determining the comprehensive prediction results, When determining the comprehensive prediction result when the results predicted by the current comments are the same, it is not necessary to obtain the confidence of the prediction model's prediction results, thereby improving the efficiency of determining the comprehensive prediction results; thirdly, when it is detected that the results predicted by each prediction model for the current comments are different, the final prediction result is determined in combination with the confidence of the prediction model's prediction results, that is, when inferring group reactions, the confidence of the prediction model's prediction results is taken into account, and the confidence of the prediction model's prediction results is at least determined by the information entropy of the probability distribution output by the prediction model, so that the confidence of the prediction model's prediction results obtained is more accurate, further improving the accuracy of the comprehensive prediction results obtained.

[0062] In addition, the ratio of the information entropy of the probability distribution output by the prediction model to the logarithmic value corresponding to the number of classifications of the prediction result of the prediction model is obtained, the difference obtained by subtracting 1 from the ratio is obtained, and the difference is used as the confidence of the prediction result of the prediction model, thereby realizing quantitative calculation of the confidence of the prediction result of the prediction model.

[0063] The category of the emotional response of the first user to the commented object or the category of the opinion response of the first user to the commented object is used as the category of the prediction result of the prediction model, thereby achieving flexible classification of the prediction result of the prediction model.

[0064] The target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or the target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models; specifically, when it is detected that the number of prediction models that predict the same result for the current comment among all prediction models is greater than half of the total number of prediction models, the prediction model that predicts the same result for the current comment is used as the initial target prediction model, and then the target prediction model is further determined according to the relationship between the first confidence of the prediction results of each initial target prediction model and the second confidence of the prediction results of the remaining prediction models, so that the final target prediction model obtained is more accurate.

[0065] Selecting an odd number of prediction models and the total number of selected prediction models being greater than a preset value improves the accuracy of the prediction results.

[0066] When determining the comprehensive prediction result, the reaction of the second user to the current comment of the first user is taken into consideration, and the weight of each comment is determined, thereby further improving the accuracy of the final prediction result.

[0067] In addition, the present invention also provides a data processing device, a computer program product and a computer-readable storage medium, which have the same or corresponding technical features as the above-mentioned data processing method and have the same effects as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0069] Figure 1 A flowchart of a data processing method provided by an embodiment of the present invention;

[0070] Figure 2 A flowchart of a method for predicting a review set using three prediction models provided by an embodiment of the present invention;

[0071] Figure 3 A structural diagram of a data processing device provided in another embodiment of the present invention. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0073] The core of the present invention is to provide a data processing method, device, product and medium to solve the technical problem that when inferring the overall response of a group of users, the use of a single model leads to low accuracy of the prediction results.

[0074] With the development of Internet technology, various social media have gradually become the mainstream tools for people's social Internet life. The huge user groups of social media express their opinions and emotions by creating, disseminating content and posting comments on social media platforms. It is of great significance to analyze the overall reaction of group users to an event on social media. The reaction of social media groups is the collection of reactions of a large number of social media users to a certain social media content, such as emotions and opinions. At present, the focus is mainly on predicting user reactions in preset and relatively limited scenarios. Using a single model, the prediction results are often biased. In addition, the group's reaction is the integration of the opinions and emotions of multiple single users, so it is necessary to gather them in a reasonable way to infer the distribution of group opinions and emotions.

[0075] Therefore, the present invention provides a method based on the fusion of confidences of multiple prediction models to solve the problem of biased prediction results of a single model at present, and integrates multiple prediction models using a prediction model fusion strategy to make unbiased predictions of social media group reactions. A method for measuring the prediction results of prediction models based on information entropy is designed, that is, using information entropy to measure the uncertainty of the prediction results of prediction models on social media group reactions, realizing the calculation of the confidence of the prediction model, and facilitating the evaluation of the prediction results of the prediction model.

[0076] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 A flowchart of a data processing method provided by an embodiment of the present invention is as follows: Figure 1 As shown, the method includes:

[0077] S10: Obtain a review set and input at least one review in the review set into multiple prediction models;

[0078] S11: respectively obtain the prediction results of each prediction model for the same comment;

[0079] S12: Determine whether the prediction results of each prediction model for the current comment are the same; if so, proceed to step S13; if not, proceed to step S14;

[0080] S13: Obtaining the average of the prediction results of all prediction models for the current comment to determine the final prediction results of multiple prediction models for the current comment;

[0081] S14: determining final prediction results of the multiple prediction models for the current comment according to the prediction result of the target prediction model for the current comment and the confidence of the prediction result of the target prediction model;

[0082] Among them, the target prediction model is a model determined from all prediction models;

[0083] S15: Determine the comprehensive prediction result of the comment set according to the sum of the final prediction results of each current comment input into the prediction model.

[0084] The comment set is the sum of comments given by different users on the comment object (such as an event). Each user can give one or more comments on the comment object. There is no restriction on the selected comment set, prediction model, etc. The selected prediction model can be used as long as it can predict the comments. The input of the prediction model is the comments, and the output is the prediction result of the comments. Representation. The social media group reaction is divided into two aspects: emotion and opinion, which are mainly shown by the content of the comments. Therefore, the prediction result can include two parts, one is the probability distribution vector of the prediction result of the comment emotion, and the other is the prediction result of the comment opinion.

[0085] There is no limit on the number of comments input into the multiple prediction models. In order to achieve a more accurate judgment of the comment set, in implementation, obtaining the comment set and inputting at least one comment in the comment set into the multiple prediction models includes:

[0086] Extract the features of the review object and the features of each review in the review set;

[0087] Obtain the features of the extracted review object and the similarity between the features of each review in the review set;

[0088] When it is detected that the similarity between the feature of the extracted comment object and the feature of the current comment in the comment set is greater than a threshold, the comment set is retained;

[0089] When it is detected that the similarity between the feature of the extracted comment object and the feature of the current comment in the comment set is less than or equal to the threshold, the current comment is removed from the comment set to obtain the comment set;

[0090] Each review in the review set is input into multiple prediction models respectively.

[0091] By filtering the comments in the comment set according to the similarity between the characteristics of the comment object and the characteristics of each comment in the comment set, and inputting each comment in the filtered comment set into multiple prediction models, a more accurate inference of the user's reaction to the comment object is achieved.

[0092] After obtaining the prediction results of each prediction model for the same comment, it can be determined that the prediction results of each prediction model for the current comment are the same. The categories of determining the prediction results of the prediction model include:

[0093] Determining in advance the reaction of the first user to the commented object based on the content of at least one first user's comment; wherein the reaction of the first user to the commented object includes at least an emotional reaction and an opinion reaction;

[0094] The category of the emotional response of the first user in the response to the commented object or the category of the opinion response of the first user in the response to the commented object is used as the category of the prediction result of the prediction model.

[0095] Furthermore, the prediction results of each prediction model for the same comment are obtained separately, including:

[0096] Obtain the categories predicted by each prediction model for the same comment and the probability distribution corresponding to the categories;

[0097] Determining whether the prediction results of each prediction model for the current comment are the same includes:

[0098] When it is detected that the categories predicted by the prediction models for the same comment are the same, determining that the prediction results of the prediction models for the current comment are the same;

[0099] When it is detected that the categories predicted by the prediction models for the same comment are different, and at least one group of prediction models predicts the same category for the same comment, it is determined that the prediction results of the prediction models for the current comment are not completely the same;

[0100] When it is detected that the categories predicted by the prediction models for the same comment are completely different, it is determined that the results predicted by the prediction models for the current comment are completely different.

[0101] When it is detected that the prediction results of the current comment by each prediction model are the same, the average of the prediction results of the current comment by all prediction models is obtained to determine the final prediction results of the current comment by multiple prediction models;

[0102] When it is detected that the prediction results of the prediction models for the current comment are different, the final prediction results of the multiple prediction models for the current comment are determined according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model. The confidence of the prediction results of the prediction models is at least determined by the information entropy of the probability distribution output by the prediction models.

[0103] The confidence of the prediction model's prediction results is determined by the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model's prediction results;

[0104] The confidence level of the prediction model's prediction results is determined based on the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model's prediction results, including:

[0105] Get the logarithmic value corresponding to the number of classifications of the prediction results of the prediction model;

[0106] Obtain the ratio of the information entropy of the probability distribution output by the prediction model to the logarithmic value corresponding to the number of classifications of the prediction results of the prediction model;

[0107] The difference between 1 and the ratio is obtained, and the difference is used as the confidence of the prediction result of the prediction model.

[0108] Since confidence helps determine the uncertainty of the model prediction results, in this embodiment, information entropy is usually used to measure the confidence of the model prediction probability, which helps model optimization. In this method, a method based on information entropy measurement is designed to predict the prediction results of the model. Referring to the existing work of sentiment analysis, the reaction of social media groups is divided into two aspects: emotions and opinions, which are mainly expressed through comments. The emotions represented by social media group comments are divided into six categories: anger, fear, happiness, love, sadness and surprise, and the opinions corresponding to the comments are divided into three categories: support, opposition and neutrality.

[0109] First, the user's comments are input into the training sentiment analysis model, and the model's sentiment prediction result for the comment is output as a probability distribution vector. .

[0110] Next, we calculate the information entropy of the above probability distribution vector The smaller the entropy, the more credible the prediction result is, and vice versa. The value range of information entropy is related to the length of the probability distribution vector, that is, ,in, is the length of the probability distribution vector, which also indicates the number of categories of the prediction result. For example, if the prediction result is an emotional category, then is 6; when the prediction result is the opinion category, is 3.

[0111] Finally, the normalization method is used to obtain the confidence score of the prediction result. The formula is as follows:

[0112] ;

[0113] The size of the confidence score reflects the credibility of the model's prediction results. The higher the score, the more credible the prediction model's prediction results are.

[0114] After determining the final prediction result of each comment in the comment set, the sum of all the final prediction results is obtained to determine the comprehensive prediction result of the combination of the comments.

[0115] In the method provided in this embodiment, firstly, compared with the method of using a single model to infer group reaction, the method provided by the present invention uses multiple prediction models to predict each comment in the comment set respectively, and determines the final prediction results of each comment by multiple prediction models, that is, inferring group reaction according to multiple prediction models, realizes the aggregation of comments of each user, and improves the accuracy of inferring group reaction; secondly, when it is detected that the results predicted by each prediction model for the current comment are the same or different, different methods are used to determine the comprehensive prediction results of the comment set, which improves the efficiency of determining the comprehensive prediction results, such as when it is detected that each prediction model predicts the same or different results for the current comment, When determining the comprehensive prediction result when the results predicted by the current comments are the same, it is not necessary to obtain the confidence of the prediction model's prediction results, thereby improving the efficiency of determining the comprehensive prediction results; thirdly, when it is detected that the results predicted by each prediction model for the current comments are different, the final prediction result is determined in combination with the confidence of the prediction model's prediction results, that is, when inferring group reactions, the confidence of the prediction model's prediction results is taken into account, and the confidence of the prediction model's prediction results is at least determined by the information entropy of the probability distribution output by the prediction model, so that the confidence of the prediction model's prediction results obtained is more accurate, further improving the accuracy of the comprehensive prediction results obtained.

[0116] In order to determine the final prediction results of multiple prediction models for the current comment, in some embodiments, when it is detected that the prediction results of the prediction models for the current comment are different, determining the final prediction results of the multiple prediction models for the current comment according to the prediction result of the target prediction model for the current comment and the confidence of the prediction result of the target prediction model includes:

[0117] When it is detected that the prediction results of the prediction models for the current comment are different, the total number of prediction models is obtained;

[0118] The target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models, or the target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models;

[0119] The final prediction results of multiple prediction models for the current comment are determined according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model.

[0120] In practice, it may be detected that the number of prediction models that predict the same result for the current comment among all prediction models is greater than half of the total number of prediction models. In this case, the target prediction model is determined based on the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models, including:

[0121] The prediction model with the same prediction result for the current comment is used as the initial target prediction model;

[0122] After determining the initial target prediction model, obtaining the first confidence of the prediction results of each initial target prediction model and the second confidence of the prediction results of the remaining prediction models; wherein the remaining prediction models are the prediction models among all the prediction models except the initial target prediction model;

[0123] Determine, according to the relationship between the second confidence level and the first confidence level, whether each remaining prediction model is used as a model in the target prediction model or not as a model in the target prediction model;

[0124] If it is determined that the remaining prediction models are models in the target prediction model, the initial target prediction model and the remaining prediction models that satisfy the relationship between the second confidence level and the first confidence level are taken as the final target prediction model.

[0125] Determining, according to the relationship between the second confidence level and the first confidence level, whether each remaining prediction model is used as a model in the target prediction model or not as a model in the target prediction model includes:

[0126] Obtain the maximum first confidence among all first confidences;

[0127] In the case where it is detected that the current second confidence is greater than the maximum first confidence, determining the remaining prediction model corresponding to the current second confidence as the model in the target prediction model;

[0128] When it is detected that the current second confidence is less than or equal to the maximum first confidence, it is determined that the remaining prediction models corresponding to the current second confidence are not used as models in the target prediction model.

[0129] In practice, it may be detected that the number of prediction models that predict the same result for the current comment among all prediction models is less than or equal to half of the total number of prediction models. In this case, determining the target prediction model based on the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models includes: taking each prediction model as the final target prediction model.

[0130] In implementation, before determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models, the method further includes:

[0131] Determine whether the total number of prediction models meets preset requirements; wherein the preset requirements at least include that the total number of prediction models is greater than or equal to a preset value, or the total number of prediction models is an odd number;

[0132] If yes, then enter the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models, or determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models;

[0133] If not, the prediction model is adjusted to enter the step of determining the target prediction model based on the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or determining the target prediction model based on the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models.

[0134] There is no limit on the preset value. Generally, the preset value is 3.

[0135] If the preset requirement is that the total number of forecast models is an odd number; adjusting the forecast model includes:

[0136] Obtain the prediction model with the lowest confidence in the prediction result among all the prediction models; eliminate the prediction model with the lowest confidence in the prediction result;

[0137] Or, obtain an odd number of new prediction models, and add the new prediction models to the existing prediction models.

[0138] If the preset requirement is that the total number of forecast models is greater than or equal to the preset value; adjusting the forecast model includes:

[0139] When the total number of detected prediction models is less than a preset value, a new prediction model is obtained;

[0140] Add new prediction models to existing prediction models.

[0141] In the above, the confidence of the prediction model prediction results is considered to determine the final prediction results of the comment set. In order to improve the accuracy of the prediction results, this embodiment further combines the weight of the comments to determine the comprehensive prediction results. The comprehensive prediction results of the comment set determined by the sum of the final prediction results of each current comment input into the prediction model include:

[0142] Obtain the second user's response to the first user's current comment;

[0143] Determine the weight of the current comment according to the reaction of the second user to the current comment of the first user;

[0144] Obtain the product of the final prediction result of each current comment and the weight of the current comment, and use the product as the new final prediction result of the current comment;

[0145] The comprehensive prediction result of the review set is determined according to the sum of the new final prediction results of each current review input into the prediction model.

[0146] In order to make those skilled in the art better understand the present invention, Figure 2 The present invention is further described in detail in the following specific embodiments. Taking the three prediction models as an example, the comprehensive prediction results of the review set are described.

[0147] Figure 2 A flowchart of a method for predicting a review set using three prediction models provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the method includes:

[0148] S16: Get a comment collection;

[0149] S17: Get comment text;

[0150] S18: Input the review text into three prediction models;

[0151] S19: Output probability distribution , and , predicted label , and ;

[0152] S20: Determine whether the three tags are the same; if so, proceed to step S21; if not, proceed to step S22;

[0153] S21: Output the mean results of three probability distributions ;

[0154] S22: Determine whether the three tags are all different; if so, proceed to step S23; if not, proceed to step S25;

[0155] S23: Obtaining information entropy and confidence score , and ;

[0156] S24: Output the results of the three probability distributions weighted by confidence scores ;

[0157] S25: Assumption Label ;

[0158] S26: Obtaining Information Entropy and confidence score , and ;

[0159] S27: Get the confidence scores of two identical labels and The larger of ;

[0160] S28: Judgment Is it greater than ; If yes, proceed to step S29; if no, proceed to step S30;

[0161] S29: Output the results of the three probability distributions weighted by confidence scores ;

[0162] S30: Output the probability distribution of two identical labels and Results weighted by confidence score ;

[0163] S31: Multiply by the number of likes plus 1;

[0164] S32: Get the final prediction label.

[0165] The prediction model in S18 can be an expert model. Specifically, the user comments are input into the prediction models of the three sentiment analysis tasks. The corresponding output category reasoning probability distributions of the three prediction model methods are , and , which correspond to the probability distribution of the three predicted categories. Assume that the category with the highest probability is the category predicted by the model. There are three cases for the above results:

[0166] (1) The prediction results of the three methods are the same;

[0167] (2) Two of the predicted results of the three methods are the same;

[0168] (3) The prediction results of the three methods are different.

[0169] For the above three situations, the present invention will design different multi-prediction model confidence fusion, the specific method is as follows:

[0170] (1) Input social media group comment collection ;

[0171] (2) For the set For each comment in the , we get the predicted category probability distribution of the three prediction models , and , predict the sentiment category of the comment based on the maximum probability in the probability distribution , and ;

[0172] (3) Based on the above analysis, the results are divided into three situations:

[0173] A. The prediction results of the three models are the same.

[0174] Final emotion category prediction probability distribution That is the mean of the prediction results of the three models, and the formula is as follows: .

[0175] B. Two of the predictions from the three models are the same.

[0176] First, calculate the probability distribution of each model output The information entropy of For the corresponding category Probability component of inference:

[0177] ;

[0178] Then the confidence score inferred by the model is calculated based on the information entropy:

[0179] ;

[0180] Of particular note is that is the number of categories, which is also the output probability distribution Length.

[0181] Afterwards, assume that the probability distributions of the two models with the same prediction results are and , and the corresponding confidence scores are and , similarly, the probability distribution and confidence score of the third different prediction result is and .

[0182] if Greater than and The maximum value of the two, then the final emotion category prediction probability distribution The weighted average of the confidence scores of the three prediction models’ prediction probabilities is:

[0183] ;

[0184] On the contrary, if Less than and The maximum value of the two, then the final emotion category prediction probability distribution The weighted average of the confidence scores of the prediction probabilities of the two prediction models with the same prediction results, that is:

[0185] ;

[0186] In this case, the predicted probability of the same result is more accurate, so there is no need to consider different prediction results.

[0187] C. The prediction results of the three models are different.

[0188] Similar to the situation in B, first calculate the information entropy of the probability distribution of each model output , and then calculate the confidence score , , The final predicted probability distribution of sentiment categories The weighted average of the confidence scores of the three prediction models’ prediction probabilities is:

[0189] ;

[0190] (4) The number of likes on comments can also represent the reaction of social media groups to a certain extent. Therefore, the present invention uses the number of likes +1 (when the number of likes for a comment is 0, 1 is the initial weight) as the weight multiplied by the probability distribution vector of the comment-induced emotions, and then they are accumulated together to obtain the social media group reaction result label of the entire comment set :

[0191] .

[0192] In the method provided in this embodiment, it is of great significance to infer the reaction of social media groups to social media groups that are different from limited scenarios such as music and movies, including the emotions of various external information on social media users and the opinions of group users on social media content; a method based on information entropy to measure the prediction results of the prediction model is designed, that is, the uncertainty of the prediction results of the prediction model on the social media group reaction is measured by using information entropy; a method based on the fusion of confidences of multiple prediction models is used to solve the problem of biased prediction results of a single prediction model at present, and a multi-prediction model fusion strategy is used to integrate multiple prediction models to make unbiased predictions on social media group reactions.

[0193] In the above embodiments, the data processing method is described in detail, and the present invention also provides corresponding embodiments of the data processing device and the data processing equipment. It should be noted that the present invention describes the embodiments of the device part from two perspectives, one is based on the functional module perspective, and the other is based on the hardware perspective.

[0194] This embodiment provides a data processing device. Based on the perspective of functional modules, this embodiment includes:

[0195] A first acquisition module, configured to acquire a review set and input at least one review in the review set into a plurality of prediction models;

[0196] The second acquisition module is used to respectively obtain the prediction results of each prediction model for the same comment;

[0197] The first determination module is used to obtain the average of the prediction results of all prediction models on the current comment to determine the final prediction results of the multiple prediction models on the current comment when it is detected that the prediction results of the current comment by each prediction model are the same; determine the comprehensive prediction result of the comment set according to the sum of the final prediction results of each current comment input into the prediction model;

[0198] The second determination module is used to determine the final prediction results of multiple prediction models for the current comment based on the prediction results of the target prediction model for the current comment and the confidence of the target prediction model's prediction results when it is detected that the prediction results of each prediction model for the current comment are different, and enter the step of determining the comprehensive prediction results of the comment set based on the sum of the final prediction results of each current comment input into the prediction model; wherein, the target prediction model is a model determined from all the prediction models; and the confidence of the prediction result of the prediction model is determined by at least the information entropy of the probability distribution output by the prediction model.

[0199] In some embodiments, the confidence of the prediction model prediction result is determined by the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model prediction result;

[0200] The data processing device includes a third determination module, which is used to determine the confidence of the prediction result of the prediction model according to the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction result of the prediction model;

[0201] The third determination module specifically includes:

[0202] The third acquisition module is used to obtain the logarithmic value corresponding to the classification number of the prediction result of the prediction model;

[0203] A fourth acquisition module is used to obtain the ratio of the information entropy of the probability distribution output by the prediction model to the logarithmic value corresponding to the number of classifications of the prediction result of the prediction model;

[0204] The module is used to obtain the difference between 1 and the ratio, and use the difference as the confidence of the prediction result of the prediction model.

[0205] In some embodiments, the data processing device includes a fourth determination module, which is used to determine the category of the prediction result of the prediction model;

[0206] The fourth determination module specifically includes:

[0207] A first determination submodule is used to determine in advance the reaction of the first user to the commented object according to the content of at least one first user's comment; wherein the reaction of the first user to the commented object includes at least an emotional reaction and an opinion reaction;

[0208] The first module is used to use the category of the emotional response of the first user to the commented object or the category of the opinion response of the first user to the commented object as the category of the prediction result of the prediction model.

[0209] In some embodiments, the second acquisition module specifically includes:

[0210] A fifth acquisition module is used to respectively obtain the categories predicted by each prediction model for the same comment and the probability distribution corresponding to the categories;

[0211] The first judgment module is used to judge whether the prediction results of each prediction model for the current comment are the same; if so, the second determination submodule is triggered, and if not, the third determination submodule is triggered;

[0212] The second determination submodule is used to determine that the prediction results of the current comment by each prediction model are the same;

[0213] The third determination submodule is used to determine that the results predicted by the prediction models for the current comment are not completely the same when it is detected that the categories predicted by the prediction models for the same comment are different, and at least one group of prediction models predicts the same category for the same comment; and to determine that the results predicted by the prediction models for the current comment are completely different when it is detected that the categories predicted by the prediction models for the same comment are completely different.

[0214] In some embodiments, the second determining module specifically includes:

[0215] A sixth acquisition module, configured to acquire the total number of prediction models when it is detected that the prediction results of the prediction models for the current comment are different;

[0216] A fourth determination submodule is used to determine the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models, or to determine the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models;

[0217] The fifth determination submodule is used to determine the final prediction results of multiple prediction models for the current comment according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model.

[0218] In some embodiments, the fourth determining submodule includes:

[0219] The second module is used to use the prediction model with the same prediction result for the current comment as the initial target prediction model when it is detected that the number of prediction models with the same prediction result for the current comment among all the prediction models is greater than half of the total number of the prediction models;

[0220] A seventh acquisition module is used to obtain the first confidence of the prediction results of each initial target prediction model and the second confidence of the prediction results of the remaining prediction models after the initial target prediction model is determined; wherein the remaining prediction models are the prediction models of all prediction models except the initial target prediction model;

[0221] a sixth determination submodule, used to determine, according to the relationship between the second confidence level and the first confidence level, whether each remaining prediction model is used as a model in the target prediction model or not as a model in the target prediction model;

[0222] The third module is used to use the initial target prediction model and the remaining prediction model that satisfies the relationship between the second confidence level and the first confidence level as the final target prediction model if it is determined that the remaining prediction model is a model in the target prediction model.

[0223] In some embodiments, the sixth determining submodule specifically includes:

[0224] An eighth acquisition module, used to acquire the maximum first confidence degree among all first confidence degrees;

[0225] A seventh determination submodule, configured to determine, when detecting that the current second confidence level is greater than the maximum first confidence level, the remaining prediction model corresponding to the current second confidence level as a model in the target prediction model;

[0226] The eighth determination submodule is used to determine that the remaining prediction models corresponding to the current second confidence level are not used as models in the target prediction model when it is detected that the current second confidence level is less than or equal to the maximum first confidence level.

[0227] In some embodiments, the fourth determination submodule is specifically used to use each prediction model as the final target prediction model when it is detected that the number of prediction models that predict the same result for the current comment among all prediction models is less than or equal to half of the total number of prediction models.

[0228] In some embodiments, the data processing apparatus further comprises:

[0229] A second judgment module is used to judge whether the total number of prediction models meets a preset requirement; wherein the preset requirement at least includes that the total number of prediction models is greater than or equal to a preset value, or the total number of prediction models is an odd number; if so, the fourth determination submodule is triggered; if not, the adjustment module is triggered;

[0230] The adjustment module is used to adjust the prediction model.

[0231] In some embodiments, the preset requirement is that the total number of prediction models is an odd number; the adjustment module is specifically used to obtain the prediction model with the smallest confidence in the prediction result among all the prediction models; and eliminate the prediction model with the smallest confidence in the prediction result;

[0232] Or, obtain an odd number of new prediction models, and add the new prediction models to the existing prediction models.

[0233] In some embodiments, the preset requirement is that the total number of prediction models is greater than or equal to a preset value; the adjustment module is specifically used to obtain a new prediction model when the total number of prediction models detected is less than the preset value; and add the new prediction model to the existing prediction model.

[0234] In some embodiments, the data processing device includes a ninth determination submodule, configured to determine a comprehensive prediction result of the review set according to the sum of the final prediction results of each current review input into the prediction model;

[0235] The ninth determination submodule specifically includes:

[0236] An eighth acquisition module, used to acquire the second user's reaction to the first user's current comment;

[0237] a tenth determination submodule, configured to determine the weight of the current comment according to the reaction of the second user to the current comment of the first user;

[0238] A ninth acquisition module, used to acquire the product of the final prediction result of each current comment and the weight of the current comment, and use the product as the new final prediction result of the current comment;

[0239] The eleventh determination submodule is used to determine the comprehensive prediction result of the comment set according to the sum of the new final prediction results of each current comment input into the prediction model.

[0240] Since the embodiments of the device part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the device part, and they will not be repeated here. The effects are the same as above.

[0241] Figure 3 This is a structural diagram of a data processing device provided by another embodiment of the present invention. This embodiment is based on the hardware perspective, such as Figure 3 As shown, the data processing device includes:

[0242] A memory 20, for storing computer programs;

[0243] The processor 21 is used to implement the steps of the data processing method mentioned in the above embodiment when executing the computer program.

[0244] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0245] The memory 20 may include one or more computer-readable storage media, which may be non-transitory. The memory 20 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 20 is at least used to store the following computer program 201, wherein, after the computer program is loaded and executed by the processor 21, it can implement the relevant steps of the data processing method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 20 may also include an operating system 202 and data 203, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 202 may include Windows, Unix, Linux, etc. The data 203 may include, but is not limited to, the data involved in the above-mentioned data processing method, etc.

[0246] In some embodiments, the data processing device may further include a display screen 22 , an input / output interface 23 , a communication interface 24 , a power supply 25 , and a communication bus 26 .

[0247] Those skilled in the art will understand that Figure 3 The structure shown in the figure does not constitute a limitation of the data processing device, and may include more or fewer components than shown in the figure.

[0248] The data processing device provided by the embodiment of the present invention includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the following method: a data processing method, with the same effect as above.

[0249] The present invention also provides a computer program product, comprising a computer program / instruction, which implements the steps of the above-mentioned data processing method when executed by a processor.

[0250] Finally, the present invention also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps recorded in the above method embodiment are implemented.

[0251] It is understandable that if the method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc. Various media that can store program codes.

[0252] The computer-readable storage medium provided by the present invention includes the above-mentioned data processing method, and the effect is the same as above.

[0253] The above is a detailed introduction to a data processing method, device, product and medium provided by the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the present invention.

[0254] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

Claims

1. A data processing method, characterized in that: include: Obtaining a review set and inputting at least one review in the review set into a plurality of prediction models; Obtain the predicted results of each prediction model for the same comment respectively; wherein the predicted results include probability distribution and predicted labels; When it is detected that the prediction results of the current comment by each prediction model are the same, the average of the prediction results of the current comment by all prediction models is obtained to determine the final prediction results of the current comment by multiple prediction models; the comprehensive prediction result of the comment set is determined according to the sum of the final prediction results of each current comment input into the prediction model; In the case where it is detected that the prediction results of the current comment by each prediction model are different, the final prediction results of the current comment by multiple prediction models are determined according to the prediction results of the current comment by the target prediction model and the confidence of the prediction results of the target prediction model, and the step of determining the comprehensive prediction results of the comment set according to the sum of the final prediction results of each current comment input into the prediction model is entered; wherein the target prediction model is a model determined from all the prediction models; the confidence of the prediction results of the prediction models is determined by at least the information entropy of the probability distribution output by the prediction model; Wherein, when it is detected that the prediction results of the prediction models for the current comment are the same, obtaining the average of the prediction results of all the prediction models for the current comment to determine the final prediction results of the multiple prediction models for the current comment includes: When it is detected according to each prediction label that the prediction results of each prediction model for the current comment are the same, the mean of the probability distribution of the prediction results of all prediction models for the current comment is obtained to determine the final prediction results of the multiple prediction models for the current comment; In the case where it is detected that the prediction results of the prediction models for the current comment are different, determining the final prediction results of the multiple prediction models for the current comment according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model includes: When it is detected that the prediction results of the prediction models for the current comment are different according to the prediction labels, the final prediction results of the multiple prediction models for the current comment are determined according to the probability distribution of the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model; Determining the target prediction model includes: The target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models, or the target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models; The step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models includes: When it is detected that the number of prediction models that predict the same result for the current comment among all the prediction models is greater than half of the total number of prediction models, the prediction model that predicts the same result for the current comment is used as the initial target prediction model; After determining the initial target prediction model, obtaining a first confidence level of the prediction results of each initial target prediction model and a second confidence level of the prediction results of the remaining prediction models; wherein the remaining prediction models are prediction models other than the initial target prediction model among all prediction models; Determining, according to the relationship between the second confidence level and the first confidence level, whether each of the remaining prediction models is to be used as a model in the target prediction model or not to be used as a model in the target prediction model; If it is determined that the remaining prediction model is a model in the target prediction model, the initial target prediction model and the remaining prediction model that satisfies the relationship between the second confidence level and the first confidence level are used as the final target prediction model; If it is determined that the remaining prediction model is not a model in the target prediction model, then determining the initial target prediction model as the final target prediction model; The confidence of the prediction model's prediction results is determined by the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model's prediction results; The confidence level of the prediction model's prediction results is determined based on the information entropy of the probability distribution output by the prediction model and the number of classifications of the prediction model's prediction results, including: Get the logarithmic value corresponding to the number of classifications of the prediction results of the prediction model; Obtain the ratio of the information entropy of the probability distribution output by the prediction model to the logarithmic value corresponding to the number of classifications of the prediction results of the prediction model; Obtaining a difference value obtained by subtracting the ratio from 1, and using the difference value as the confidence level of the prediction result of the prediction model; The step of determining, according to the relationship between the second confidence level and the first confidence level, whether each of the remaining prediction models is used as a model in the target prediction model or not as a model in the target prediction model comprises: Obtaining the maximum first confidence degree among all the first confidence degrees; In the case where it is detected that the current second confidence is greater than the maximum first confidence, determining the remaining prediction model corresponding to the current second confidence as a model in the target prediction model; When it is detected that the current second confidence is less than or equal to the maximum first confidence, it is determined that the remaining prediction model corresponding to the current second confidence is not used as a model in the target prediction model.

2. The data processing method according to claim 1, characterized in that: The categories that determine the prediction results of the forecast model include: Determining in advance the reaction of the first user to the commented object based on the content of at least one first user's comment; wherein the reaction of the first user to the commented object includes at least an emotional reaction and an opinion reaction; The category of the emotional response of the first user in the response to the commented object or the category of the opinion response of the first user in the response to the commented object is used as the category of the prediction result of the prediction model.

3. The data processing method according to claim 2, characterized in that: The obtaining of the prediction results of each prediction model for the same comment includes: Obtain the categories predicted by each prediction model for the same comment and the probability distribution corresponding to the categories; Determining whether the prediction results of each prediction model for the current comment are the same includes: When it is detected that the categories predicted by the prediction models for the same comment are the same, determining that the prediction results of the prediction models for the current comment are the same; When it is detected that the categories predicted by the prediction models for the same comment are different, and at least one group of prediction models predicts the same category for the same comment, it is determined that the prediction results of the prediction models for the current comment are not completely the same; When it is detected that the categories predicted by the prediction models for the same comment are completely different, it is determined that the results predicted by the prediction models for the current comment are completely different.

4. The data processing method according to claim 3, characterized in that: When it is detected that the prediction results of the prediction models for the current comment are different, the final prediction results of the multiple prediction models for the current comment are determined according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model, including: When it is detected that the prediction results of the prediction models for the current comment are different, the total number of prediction models is obtained; The target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models, or the target prediction model is determined according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models and the confidence of the prediction results of all the prediction models; The final prediction results of multiple prediction models for the current comment are determined according to the prediction results of the target prediction model for the current comment and the confidence of the prediction results of the target prediction model.

5. The data processing method according to claim 1, characterized in that: Determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all the prediction models and the total number of prediction models includes: When it is detected that the number of prediction models that predict the same result for the current comment among all the prediction models is less than or equal to half of the total number of prediction models, each prediction model is used as the final target prediction model.

6. The data processing method according to claim 4, characterized in that: Before determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models, the method further includes: Determine whether the total number of prediction models meets preset requirements; wherein the preset requirements at least include that the total number of prediction models is greater than or equal to a preset value, or that the total number of prediction models is an odd number; If yes, then enter the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models, or determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models; If not, the prediction model is adjusted and the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models is entered, or the step of determining the target prediction model according to the relationship between the number of prediction models that predict the same result for the current comment among all prediction models and the total number of prediction models and the confidence of the prediction results of all prediction models is entered.

7. The data processing method according to claim 6, characterized in that: The preset requirement is that the total number of prediction models is an odd number; adjusting the prediction model includes: Obtain the prediction model with the lowest confidence in the prediction result among all the prediction models; eliminate the prediction model with the lowest confidence in the prediction result; Or, obtain an odd number of new prediction models, and add the new prediction models to the existing prediction models.

8. The data processing method according to claim 6, characterized in that: The preset requirement is that the total number of prediction models is greater than or equal to a preset value; adjusting the prediction model includes: When the total number of detected prediction models is less than the preset value, obtaining a new prediction model; Add new prediction models to existing prediction models.

9. The data processing method according to any one of claims 1 to 8, characterized in that: The step of determining the comprehensive prediction result of the review set according to the sum of the final prediction results of each current review input into the prediction model includes: Obtain the second user's response to the first user's current comment; Determine the weight of the current comment according to the reaction of the second user to the current comment of the first user; Obtain the product of the final prediction result of each current comment and the weight of the current comment, and use the product as the new final prediction result of the current comment; The comprehensive prediction result of the review set is determined according to the sum of the new final prediction results of each current review input into the prediction model.

10. A data processing device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the data processing method according to any one of claims 1 to 9 when executing the computer program.

11. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the data processing method according to any one of claims 1 to 9 are implemented.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the data processing method according to any one of claims 1 to 9 are implemented.

Images