Data analysis method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a data analysis method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining a to-be-processed text, and performing word segmentation processing on the to-be-processed text to obtain a word segmentation data set; performing analysis processing on the word segmentation data in the word segmentation data set through a hybrid expert model, and determining a weight value corresponding to each expert network in the hybrid expert model based on an analysis processing result and a gating network; performing sorting processing and cumulative probability calculation on the weight values corresponding to the expert networks, and determining the number of the expert networks and a target expert network based on a cumulative probability calculation result; and based on a target expert network of the expert network number and a weight value corresponding to the target expert network, determining an analysis decision result of the word segmentation data until analysis decision results of all word segmentation data in the word segmentation data set are obtained, and determining a data analysis result of the to-be-processed text. By adopting the method, the overall decision-making accuracy of the hybrid expert model is improved.

Description

Technical Field

[0001] The present application relates to the field of computer science and artificial intelligence technology, and in particular to a data analysis method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] Mixture of Experts (MoE) is a deep learning framework that handles complex data analysis problems by integrating the decisions of multiple "experts" in the mixed expert model. In the mixed expert model, the gating network plays a key role in this deep learning framework. It adjusts the weights according to the current input and determines the influence of each expert network in the final decision.

[0003] In traditional technology, sentiment discrimination or semantic analysis of text data is usually performed through a hybrid expert model. Taking semantic analysis of text data as an example, the number of experts involved in the decision-making is preset, that is, the K value (K is an integer greater than 1). Then, according to the weights assigned by the gated network in the hybrid expert model and the preset number of experts (K value), the target expert network involved in the text data decision-making is determined in the hybrid expert model, and then the semantic analysis decision of the text data is performed based on the K target expert networks to obtain the semantic analysis results of the text data.

[0004] However, in traditional technology, the pre-set K value makes the mixed expert model less flexible. When the K value is too large, it is easy to waste computing resources. When the K value is too small, the accuracy of the model analysis decision results is poor. Therefore, a data analysis method with a reasonable setting of the K value is proposed. Summary of the invention

[0005] Based on this, it is necessary to provide a data analysis method, apparatus, computer equipment, computer-readable storage medium and computer program product to address the above technical problems.

[0006] In a first aspect, the present application provides a data analysis method, comprising:

[0007] Obtaining a text to be processed, and performing word segmentation processing on the text to be processed to obtain a word segmentation data set;

[0008] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, a weight value corresponding to each expert network in the hybrid expert model is determined;

[0009] Sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results;

[0010] Based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0011] In one embodiment, for each word segmentation data in the word segmentation data set, analyzing and processing the word segmentation data through a hybrid expert model, and determining the weight value corresponding to each expert network in the hybrid expert model based on the analysis and processing result and the gating network, including:

[0012] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model to obtain an analysis and processing result;

[0013] Based on the analysis and processing results and the gating network in the hybrid expert model, the weight values ​​corresponding to each expert network in the hybrid expert model are calculated; the weight values ​​corresponding to the expert networks are used to characterize the decision-making efficiency of the expert networks on the word segmentation data.

[0014] In one embodiment, the weight values ​​corresponding to the expert networks are sorted and the cumulative probability is calculated, and the number of expert networks and the target expert network are determined based on the cumulative probability calculation result, including:

[0015] Normalizing the weight values ​​corresponding to each of the expert networks to obtain probability weight values ​​corresponding to each of the expert networks;

[0016] Sorting the probability weight values ​​corresponding to the expert networks in descending order to obtain a weight value sequence;

[0017] Performing cumulative probability calculation on the probability weight values ​​in the weight value sequence, and verifying the cumulative probability calculation result after each probability weight value is accumulated based on a cumulative probability threshold;

[0018] If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of expert networks and the target expert network are determined.

[0019] In one embodiment, if the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, determining the number of expert networks and the target expert network includes:

[0020] If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of probability weight values ​​included in the current cumulative probability calculation result is determined as the number of expert networks;

[0021] The expert network corresponding to the probability weight value included in the current cumulative probability calculation result is determined as the target expert network.

[0022] In one embodiment, the method further comprises:

[0023] If the cumulative probability calculation result of all probability weight values ​​in the weight value sequence is less than the cumulative probability threshold, the number of expert networks in the hybrid expert model is determined as the number of expert networks, and each expert network in the hybrid expert model is determined as a target expert network.

[0024] In one embodiment, the target expert network based on the number of expert networks and the weight value corresponding to the target expert network determines the analysis and decision result of the word segmentation data, including:

[0025] Analyze and make decisions on the word segmentation data through the target expert network in the hybrid expert model to obtain an initial analysis and decision result;

[0026] Based on the initial analysis and decision results of each of the target expert networks and the weight values ​​corresponding to the target expert networks, the analysis and decision results of the word segmentation data are determined.

[0027] In a second aspect, the present application also provides a data analysis device, comprising:

[0028] An acquisition module is used to acquire the text to be processed and perform word segmentation processing on the text to be processed to obtain a word segmentation data set;

[0029] A first determination module is used to analyze and process each word segmentation data in the word segmentation data set through a hybrid expert model, and determine a weight value corresponding to each expert network in the hybrid expert model based on the analysis and processing result and the gating network;

[0030] A second determination module is used to sort the weight values ​​corresponding to the expert networks and calculate the cumulative probability, and determine the number of expert networks and the target expert network based on the cumulative probability calculation result;

[0031] The third determination module is used to determine the analysis and decision results of the word segmentation data based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0032] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0033] Obtaining a text to be processed, and performing word segmentation processing on the text to be processed to obtain a word segmentation data set;

[0034] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, a weight value corresponding to each expert network in the hybrid expert model is determined;

[0035] Sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results;

[0036] Based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0038] Obtaining a text to be processed, and performing word segmentation processing on the text to be processed to obtain a word segmentation data set;

[0039] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, a weight value corresponding to each expert network in the hybrid expert model is determined;

[0040] Sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results;

[0041] Based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0042] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0043] Obtaining a text to be processed, and performing word segmentation processing on the text to be processed to obtain a word segmentation data set;

[0044] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, a weight value corresponding to each expert network in the hybrid expert model is determined;

[0045] Sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results;

[0046] Based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0047] The above-mentioned data analysis method, device, computer equipment, computer-readable storage medium and computer program product obtain a text to be processed, and perform word segmentation processing on the text to be processed to obtain a word segmentation data set; for each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, the weight value corresponding to each expert network in the hybrid expert model is determined; the weight value corresponding to each expert network is sorted and cumulative probability is calculated, and based on the cumulative probability calculation result, the number of expert networks and the target expert network are determined; based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis decision result of the word segmentation data is determined, until the analysis decision result of all the word segmentation data in the word segmentation data set is obtained, and the data analysis result of the text to be processed is determined. This method is used to process the word segmentation data in the text to be processed, refine the text processing granularity, and dynamically adjust the number of expert networks and the target expert network corresponding to each word segmentation data by accumulating the probability calculation of the weight value of each expert network. This avoids the waste of computing resources when the number of expert networks is too large, and also avoids the poor decision accuracy when the number of expert networks is small, thereby improving the overall decision accuracy of the hybrid expert model. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A schematic diagram of a data analysis method in one embodiment;

[0050] Figure 2 A schematic diagram of a flow chart of a step of calculating weight values ​​corresponding to each expert network in one embodiment;

[0051] Figure 3 A schematic diagram of a flow chart of steps for determining the number of expert networks and target expert networks in one embodiment;

[0052] Figure 4 A flowchart of the steps of obtaining expert network data and a target expert network in one embodiment;

[0053] Figure 5 A schematic diagram of a flow chart of steps for determining a target expert network under special circumstances in one embodiment;

[0054] Figure 6 A schematic diagram of a flow chart of steps for determining an analysis and decision result of word segmentation data in one embodiment;

[0055] Figure 7 A schematic diagram of a specific example flow of a process for determining the number of expert networks and the target expert network in one embodiment;

[0056] Figure 8 A schematic diagram of a specific process of performing a data analysis task of a text to be processed based on a hybrid expert model in one embodiment;

[0057] Fig. 9 is a structural block diagram of a data analysis device in one embodiment;

[0058] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0060] In one embodiment, the traditional hybrid expert model MoE adopts the Top-K strategy to select experts, that is, the first K experts with the highest weights are selected according to the weights assigned by the gating network to participate in the final decision. However, for the traditional hybrid expert model, the determined K value is the same value for the analysis task of the entire text data. For example, K is pre-defined as 4 in the configuration file of the model. Then, before the hybrid expert model is started, K=4 is read from the configuration file to start the hybrid expert model. Once the model is started, the K value cannot be modified for the analysis and processing of the entire text data. In this way, when the K value is set too large, for example, 3 expert models can determine an accurate decision analysis result, and 4 expert models are used for decision analysis, which easily leads to a waste of computing resources. When the K value is set too small, for example, 6 expert models are required to participate in the decision analysis, then only 4 expert networks are used for decision analysis, and the accuracy of the model analysis decision result is poor. Therefore, a data analysis method for reasonably setting the K value is continued. Figure 1 As shown, the present application provides a data analysis method. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0061] Step 102, obtaining the text to be processed, and performing word segmentation processing on the text to be processed to obtain a word segmentation data set.

[0062] In implementation, a hybrid expert model is pre-integrated in the terminal, and the hybrid expert model can be used to perform data analysis on the text data of the text to be processed. The specific data analysis task can be a semantic analysis task or a sentiment discrimination task. The embodiment of the present application does not limit the data analysis tasks that can be performed by the hybrid expert model. In this way, when the terminal executes a data analysis task on a certain text data of the text to be processed, the terminal first obtains the text to be processed, and then, based on the pre-model layer of the hybrid expert model, or based on other text processing models, performs word segmentation processing on the text to be processed to obtain a word segmentation data set corresponding to the text to be processed. The word segmentation data set contains multiple word segmentation data tokens after text segmentation.

[0063] Step 104: for each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by the hybrid expert model, and based on the analysis and processing results and the gating network, the weight value corresponding to each expert network in the hybrid expert model is determined.

[0064] In implementation, for each word segmentation data token in the word segmentation data set, the terminal performs initial analysis and processing on the word segmentation data through each expert network included in the hybrid expert model, and obtains the preliminary analysis and processing results given by each expert network. For example, the analysis and processing result can be the semantic analysis probability or the emotion discrimination probability for the word segmentation data. Then, the gating network in the hybrid expert model calculates the corresponding weight value of each expert network for the process of each expert network outputting the corresponding analysis and processing result. Each weight value is used to reflect the decision-making efficiency of the expert network for the input word segmentation data.

[0065] Step 106 , sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results.

[0066] In the implementation, for the weight values ​​corresponding to each determined expert network, the terminal sorts the weight values ​​corresponding to each expert network and calculates the cumulative probability. Then, the terminal selects the first N expert networks to participate in the analysis and decision of this round of word segmentation data through a preset cumulative probability threshold (such as 90% or other user-defined values) and the calculated cumulative probability result, that is, the terminal determines the number of expert networks and the target expert network corresponding to the analysis and decision of the word segmentation data. It can be understood that for different word segmentation data, the cumulative probability results determined are different, and the number N of expert networks corresponding to the word segmentation data may be different.

[0067] Step 108, based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, determine the analysis and decision results of the word segmentation data, until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0068] In implementation, the terminal performs analysis and decision processing on the word segmentation data based on the determined number of expert networks and the target expert networks, and performs weighted processing on the analysis and decision results output by each target expert network in combination with the weight values ​​corresponding to each target expert network, to obtain the analysis and decision results of the word segmentation data. In this way, the hybrid expert model performs the processing process of steps 104 to 106 on each word segmentation data set until the analysis and decision results of all word segmentation data in the word segmentation data set are obtained. In the process of obtaining the analysis and decision results of all word segmentation data, after each analysis and decision of a word segmentation data is executed, the hybrid expert model learns the analysis and decision results of the corresponding word segmentation data. After obtaining the analysis and decision results of all word segmentation data, the hybrid expert model has learned the correlation between the analysis and decision results of all word segmentation data. The terminal performs a comprehensive analysis on the analysis and decision results of all word segmentation data to determine the data analysis results of the text to be processed.

[0069] In the above data analysis method, a text to be processed is obtained, and the text to be processed is segmented to obtain a segmentation data set; for each segmentation data in the segmentation data set, the segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, the weight values ​​corresponding to each expert network in the hybrid expert model are determined; the weight values ​​corresponding to each expert network are sorted and cumulatively calculated, and the number of expert networks and the target expert network are determined based on the cumulative probability calculation results; the analysis and decision results of the segmentation data are determined based on the target expert network of the number of expert networks and the weight values ​​corresponding to the target expert network, until the analysis and decision results of all the segmentation data in the segmentation data set are obtained, and the data analysis results of the text to be processed are determined. By processing the segmentation data in the text to be processed, the text processing granularity is refined, and by performing cumulative probability calculation on the weight values ​​of each expert network, the number of expert networks corresponding to each segmentation data and the target expert network are dynamically adjusted, thereby avoiding the waste of computing resources when the number of expert networks is too large, and avoiding the poor decision accuracy when the number of expert networks is small, thereby improving the overall decision accuracy of the hybrid expert model.

[0070] In an exemplary embodiment, Figure 2 As shown, the specific processing process of step 104 includes the following steps 202 to 204. Among them:

[0071] Step 202: for each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by using a hybrid expert model to obtain an analysis and processing result.

[0072] In implementation, for each word segmentation data token in the word segmentation data set, the terminal analyzes and processes the word segmentation data token through a hybrid expert model to obtain the analysis and processing results of the word segmentation data by each expert network in the hybrid expert model. Corresponding to different data analysis tasks, the meaning represented by the analysis and processing results may be different. For example, for the emotion discrimination task of text data, the analysis and processing result can be the emotion discrimination result of each expert network on the word segmentation data token, such as sad, happy, positive, negative, etc. If it is a semantic analysis task for text data, the analysis and processing result can be the semantic analysis result of each expert network on the word segmentation data token. For example, the semantic analysis result can determine whether A (name) executes event B, etc. This application does not limit the data analysis tasks that can be performed by the hybrid expert model on natural language text data.

[0073] Step 204, based on the analysis and processing results and the gating network in the hybrid expert model, the weight value corresponding to each expert network in the hybrid expert model is calculated.

[0074] Among them, the weight value corresponding to the expert network is used to characterize the decision-making efficiency of the expert network for word segmentation data.

[0075] In implementation, in addition to the expert networks, the hybrid expert model also includes a gating network, which is a key part of the hybrid expert model and is used to enable the model to learn to select the type of subtask for a given input and to select a trusted expert network for strong predictions. Therefore, the gating network screens the expert networks by setting the weight values ​​of each expert network. The terminal processes the features of the current word segmentation data based on the hybrid expert model and calculates the weight values ​​corresponding to each expert network in the hybrid expert model. The weight values ​​obtained reflect the relative efficiency (decision-making effectiveness) of different experts on the input token.

[0076] In this embodiment, the weight values ​​corresponding to each expert network in the hybrid expert model are calculated through a gated network. Then, based on the weight values ​​corresponding to each expert network and the preset cumulative probability calculation method, the number of expert networks corresponding to the current word segmentation data and the corresponding target expert network can be determined, thereby realizing dynamic adjustment of the number of expert networks in the data analysis task of the processed text, improving the flexibility of data analysis task processing, and improving the quality of data analysis decision-making.

[0077] In an exemplary embodiment, Figure 3 As shown, step 106 includes steps 302 to 306. Among them:

[0078] Step 302, normalizing the weight values ​​corresponding to each expert network to obtain the probability weight values ​​corresponding to each expert network.

[0079] In the implementation, for the weight values ​​of each expert network determined by the gated network, the terminal performs a normalization process on each weight value. Specifically, the terminal can normalize the weight values ​​corresponding to each expert network through a single softmax function or multiple softmax functions (multiple softmax functions can further distinguish the degree of sharing of experts) to obtain the probability weight value corresponding to each expert network. The weights corresponding to all expert networks are converted into a probability distribution to ensure that the sum of the probability weight values ​​corresponding to all expert networks is 1.

[0080] Step 304, sorting the probability weight values ​​corresponding to each expert network in descending order to obtain a weight value sequence.

[0081] In the implementation, the terminal sorts the normalized probability weight values ​​from large to small to obtain a weight value sequence containing the probability weight values ​​corresponding to each expert network. For example, the hybrid expert model contains 5 expert networks, and the probability weight values ​​corresponding to these 5 expert networks are sorted from high to low as [0.4, 0.3, 0.2, 0.05, 0.05], and then the weight value sequence corresponding to the 5 expert networks is obtained.

[0082] Step 306, performing cumulative probability calculation on the probability weight values ​​in the weight value sequence, and verifying the cumulative probability calculation result after each probability weight value is accumulated based on the cumulative probability threshold.

[0083] In implementation, the terminal performs cumulative probability calculation on the probability weight values ​​in the weight value sequence, that is, according to the order of the probability weight values ​​in the weight value sequence from large to small, each probability weight value is accumulated one by one, and each time a probability weight value is accumulated, a new cumulative probability calculation result is obtained. At this time, the cumulative probability calculation result needs to be verified, that is, the size relationship between the current cumulative probability calculation result and the cumulative probability threshold is compared to determine whether to continue to accumulate the probability weight value. For example, when processing a word segmentation data token, the probability weight values ​​of the five experts are [0.4, 0.3, 0.2, 0.05, 0.05] from high to low, and the cumulative probability threshold P is set in advance. The cumulative probability threshold P=0.6, then when the probability weight values ​​of the five experts are cumulatively calculated in the order of the probability weight values ​​in the weight value sequence from high to low, the first probability weight value is 0.4, 0.4<0.6, then the probability weight values ​​continue to be accumulated, and the second probability weight value is accumulated, (0.4+0.3)>0.6, then the probability weight values ​​are stopped.

[0084] Step 308: If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of expert networks and the target expert network are determined.

[0085] In implementation, if the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the terminal determines the number of expert networks and the target expert network for the current segmentation data token. Specifically, as in the specific processing process in step 306 above, in the process of cumulative probability calculation of the probability weight value, each time a probability weight value is accumulated, the probability weight value is verified based on the cumulative probability threshold. If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, such as (0.4+0.3)>0.6 in the above example, the terminal stops accumulating the probability weight value, that is, stops selecting the expert network, and uses the number of currently accumulated expert networks as the expert network data, and uses the expert network corresponding to the currently accumulated probability weight value as the target expert network. The cumulative probability threshold (for example, 0.6) is used to determine the selection of the first N experts (N is different for different tokens, for example, N=2 in this example), so that their cumulative weights at least reach this cumulative probability threshold. This ensures that the selected expert set can represent the vast majority of the "opinions" of the entire expert pool on the current input data.

[0086] In this embodiment, by calculating the cumulative probability of the probability weight values ​​corresponding to each expert network and verifying the preset cumulative probability threshold, the number of expert networks corresponding to the current word segmentation data and the corresponding target expert network are dynamically determined, thereby improving the decision quality.

[0087] In an exemplary embodiment, Figure 4 As shown, the specific processing process of step 308 includes:

[0088] Step 401: If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of probability weight values ​​included in the current cumulative probability calculation result is determined as the number of expert networks.

[0089] In implementation, if the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the terminal determines the number of probability weight values ​​contained in the current cumulative probability calculation result as the number of expert networks. For example, in the process of accumulating probability weight values, the current cumulative probability calculation result is the result of accumulating the first three probability weight values ​​in the weight value sequence, and the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, then the terminal determines the number of probability weight values ​​contained in the cumulative probability calculation result, i.e., 3, as the number of expert networks selected in the hybrid expert model.

[0090] Step 402: determine the expert network corresponding to the probability weight value contained in the current cumulative probability calculation result as the target expert network.

[0091] In implementation, since the probability weight values ​​are sorted from large to small in the weight value sequence, and the probability weight values ​​are also used to reflect the decision-making efficiency of the expert network for the word segmentation data, the probability weight values ​​from large to small in the weight value sequence are also used to reflect the decision-making efficiency of the expert network for the word segmentation data from large to small. The expert network with greater decision efficiency has a greater influence on the decision analysis of the word segmentation data, and the expert network with smaller decision efficiency has a smaller influence on the decision analysis of the word segmentation data. Therefore, after determining the number of expert networks N selected in the hybrid expert model, for example, N=3, the expert networks corresponding to the first three probability weight values ​​are the target expert networks selected to participate in the decision of the word segmentation data.

[0092] In this embodiment, the target expert network is selected in the hybrid expert model through the cumulative probability calculation results and the preset cumulative probability threshold, so that the target expert network can participate in the decision-making. By verifying the cumulative probability threshold, the decision opinion given by the selected target expert network can represent the decision opinions of all expert networks in the hybrid expert model. In addition, the verification of the cumulative probability threshold is performed in the process of accumulating the probability weight values ​​one by one, and the number of selected target expert networks will not be redundant, thereby avoiding waste of computing resources.

[0093] In an exemplary embodiment, for a special case, such as Figure 5 As shown, the method also includes:

[0094] Step 501, if the cumulative probability calculation result of all probability weight values ​​in the weight value sequence is less than the cumulative probability threshold, the number of expert networks in the hybrid expert model is determined as the number of expert networks, and each expert network in the hybrid expert model is determined as a target expert network.

[0095] In implementation, in the process of accumulating the probability weight values ​​corresponding to each expert network in the hybrid expert model, if after accumulating all the probability weight values ​​in the weight value sequence, the cumulative probability calculation result is still less than the preset cumulative probability threshold, for example, the preset cumulative probability threshold is relatively large P=0.9, then it may be necessary to select all the expert networks in the hybrid expert model to meet the threshold condition, that is, the terminal determines the number of expert networks in the hybrid expert model as the number of expert networks. The terminal determines each expert network in the hybrid expert model as the target expert network.

[0096] In this embodiment, when the cumulative probability calculation result after adding up all the probability weight values ​​in the weight value sequence is still less than the preset cumulative probability threshold, all expert networks are selected to participate in the decision-making of the word segmentation data to ensure the accuracy of the decision corresponding to the word segmentation data.

[0097] In an exemplary embodiment, Figure 6As shown, in step 108, based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the specific processing process of determining the analysis and decision result of the word segmentation data includes:

[0098] Step 602: Analyze and make decisions on the word segmentation data through the target expert network in the hybrid expert model to obtain an initial analysis and decision result.

[0099] In implementation, based on the determination of the number of expert networks and the target expert network for each word segmentation data by the hybrid expert model in the above embodiment, a corresponding number of target expert networks in the hybrid expert model analyze and make decisions on each word segmentation data to obtain the initial analysis and decision results corresponding to the word segmentation data by each target expert network.

[0100] Step 604: Determine the analysis and decision results of the word segmentation data based on the initial analysis and decision results of each target expert network and the weight value corresponding to the target expert network.

[0101] In the implementation, the terminal determines the analysis and decision results of the word segmentation data based on the initial analysis and decision results of each target expert network and the weight value corresponding to the target expert network. Specifically, a weighted calculation is performed on the initial analysis and decision results of the word segmentation data given by each target expert network and the weight value corresponding to the target expert network to obtain the final analysis and decision result of the word segmentation data.

[0102] Optionally, in the process of the hybrid expert model performing the data analysis task on the text to be processed, the hybrid expert model uses the attention mechanism to analyze which parts of the input text to be processed are more important. The attention mechanism interacts with the gating network and assigns different weights to different parts of the text to be processed accordingly. When the hybrid expert model makes the final decision, the hybrid expert model also strengthens the weight of each word segmentation data in the text to be processed through the attention mechanism. In this way, after obtaining the analysis and decision results corresponding to each word segmentation data of the text to be processed, based on the attention mechanism in the hybrid expert model, the final decision on the text to be processed can be completed, that is, the data analysis result of the text to be processed can be obtained.

[0103] In this embodiment, the number of expert networks and the target expert network are dynamically selected for the word segmentation data in the text to be processed based on the hybrid expert model, and analysis and decision making are performed on each word segmentation data in the text to be processed based on the selected target expert network, thereby improving the decision quality of the hybrid expert model and ensuring the accuracy of the decision corresponding to each word segmentation data.

[0104] In an optional embodiment, the final prediction output is integrated according to the selected target expert network output and the weight corresponding to the target expert network. According to the performance of the actual hybrid expert model in the application, the weight of the gating network is adjusted to optimize the cumulative probability threshold setting, or the gating mechanism is adaptively adjusted through feedback and learning process.

[0105] In an optional embodiment, if Figure 7 As shown, a specific example process of determining the number of expert networks and the target expert network is provided, and the example process includes:

[0106] Step 701, the gating network calculates the weight value corresponding to each expert network in the hybrid expert model;

[0107] Step 702, normalizing the weight values ​​corresponding to each expert network to obtain the probability weight values ​​corresponding to each expert network;

[0108] Step 703, sorting the probability weight values ​​corresponding to each expert network in descending order to obtain a weight value sequence;

[0109] Step 704, calculating the cumulative probability calculation result after accumulating each probability weight value in the weight value sequence;

[0110] Step 705, verifying the cumulative probability calculation result after accumulating each probability weight value, if the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, executing step 707, if the current cumulative probability calculation result is less than the cumulative probability threshold, executing step 706;

[0111] Step 706, determine whether all probability weight values ​​in the weight sequence have been accumulated, that is, whether all expert networks have been considered. If so, execute step 707, if not, accumulate the next probability weight value and execute step 704;

[0112] Step 707, stop selecting the expert network;

[0113] Step 708, determining the number of probability weight values ​​included in the current cumulative probability calculation result as the number of expert networks, and determining the expert network corresponding to the probability weight value included in the current cumulative probability calculation result as the target expert network;

[0114] Step 709: Determine the analysis and decision result of the word segmentation data based on the target expert network and the weight value corresponding to the target expert network.

[0115] In an optional embodiment, if Figure 8 As shown, a specific example process for performing a data analysis task of a text to be processed based on a hybrid expert model is provided, and the example process includes:

[0116] Step 801, performing word segmentation processing on the text to be processed, obtaining a word segmentation data set and inputting it into the hybrid expert model;

[0117] Step 802, strengthening the weight of each word segmentation data in the text to be processed through the attention mechanism;

[0118] Step 803, the weighted word segmentation data is then transmitted to the gated network, and the gated network calculates the weight values ​​corresponding to each expert network in the hybrid expert model;

[0119] Step 804, normalizing the weight values ​​corresponding to each expert network to obtain the probability weight values ​​corresponding to each expert network;

[0120] Step 805, sorting the probability weight values ​​corresponding to each expert network in descending order to obtain a weight value sequence;

[0121] Step 806, calculating the cumulative probability calculation result after accumulating each probability weight value in the weight value sequence;

[0122] Step 807, verifying the cumulative probability calculation result after accumulating each probability weight value, if the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, executing step 809, if the current cumulative probability calculation result is less than the cumulative probability threshold, executing step 808;

[0123] Step 808, determine whether all probability weight values ​​in the weight sequence have been accumulated, that is, whether all expert networks have been considered. If so, execute step 809, if not, accumulate the next probability weight value and execute step 806;

[0124] Step 809, determining the number of expert networks and the target expert network based on the cumulative probability calculation result and the cumulative probability threshold;

[0125] Step 810, determining the analysis and decision result of the word segmentation data based on the target expert network and the weight value corresponding to the target expert network;

[0126] Step 811, until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, the data analysis results of the text to be processed are output.

[0127] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0128] Based on the same inventive concept, the embodiment of the present application also provides a data analysis device for implementing the data analysis method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more data analysis device embodiments provided below can refer to the limitations on the data analysis method above, and will not be repeated here.

[0129] In an exemplary embodiment, Fig. 9 As shown, a data analysis device is provided, including: an acquisition module 901, a first determination module 902, a second determination module 903 and a third determination module 904, wherein:

[0130] The acquisition module 901 is used to acquire the text to be processed and perform word segmentation processing on the text to be processed to obtain a word segmentation data set;

[0131] The first determination module 902 is used to analyze and process each word segmentation data in the word segmentation data set through the hybrid expert model, and determine the weight value corresponding to each expert network in the hybrid expert model based on the analysis and processing result and the gating network;

[0132] The second determination module 903 is used to sort the weight values ​​corresponding to each expert network and calculate the cumulative probability, and determine the number of expert networks and the target expert network based on the cumulative probability calculation result;

[0133] The third determination module 904 is used to determine the analysis and decision results of the word segmentation data based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0134] In one embodiment, the first determination module 902 is specifically used to analyze and process each segmentation data in the segmentation data set by using a hybrid expert model to obtain an analysis and processing result;

[0135] Based on the analysis and processing results and the gating network in the hybrid expert model, the weight values ​​corresponding to each expert network in the hybrid expert model are calculated; the weight values ​​corresponding to the expert networks are used to characterize the decision-making efficiency of the expert networks for word segmentation data.

[0136] In one embodiment, the second determination module 903 is specifically used to normalize the weight values ​​corresponding to each expert network to obtain the probability weight values ​​corresponding to each expert network;

[0137] Sort the probability weight values ​​corresponding to each expert network in descending order to obtain a weight value sequence;

[0138] Perform cumulative probability calculation on the probability weight values ​​in the weight value sequence, and verify the cumulative probability calculation result after adding up each probability weight value based on the cumulative probability threshold;

[0139] If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of expert networks and the target expert network are determined.

[0140] In one embodiment, the second determination module 903 is specifically configured to determine the number of expert networks and the target expert network if the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, including:

[0141] If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of probability weight values ​​contained in the current cumulative probability calculation result is determined as the number of expert networks;

[0142] The expert network corresponding to the probability weight value contained in the current cumulative probability calculation result is determined as the target expert network.

[0143] In one embodiment, the device further comprises:

[0144] If the cumulative probability calculation result of all probability weight values ​​in the weight value sequence is less than the cumulative probability threshold, the number of expert networks in the hybrid expert model is determined as the number of expert networks, and each expert network in the hybrid expert model is determined as the target expert network.

[0145] In one embodiment, the third determination module 904 is specifically used to analyze and make decisions on the word segmentation data through the target expert network in the hybrid expert model to obtain an initial analysis and decision result;

[0146] Based on the initial analysis and decision results of each target expert network and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined.

[0147] Each module in the above data analysis device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0148] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Fig.10 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a data analysis method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0149] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0150] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0151] Obtain the text to be processed, and perform word segmentation on the text to be processed to obtain a word segmentation dataset;

[0152] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by the hybrid expert model, and based on the analysis and processing results and the gating network, the weight value corresponding to each expert network in the hybrid expert model is determined;

[0153] The weight values ​​corresponding to each expert network are sorted and the cumulative probability is calculated, and based on the cumulative probability calculation results, the number of expert networks and the target expert network are determined;

[0154] Based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

[0155] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0156] For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by the hybrid expert model to obtain the analysis and processing results;

[0157] Based on the analysis and processing results and the gating network in the hybrid expert model, the weight values ​​corresponding to each expert network in the hybrid expert model are calculated; the weight values ​​corresponding to the expert networks are used to characterize the decision-making efficiency of the expert networks for word segmentation data.

[0158] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0159] Normalize the weight values ​​corresponding to each expert network to obtain the probability weight values ​​corresponding to each expert network;

[0160] Sort the probability weight values ​​corresponding to each expert network in descending order to obtain a weight value sequence;

[0161] Perform cumulative probability calculation on the probability weight values ​​in the weight value sequence, and verify the cumulative probability calculation result after adding up each probability weight value based on the cumulative probability threshold;

[0162] If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of expert networks and the target expert network are determined.

[0163] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0164] If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of probability weight values ​​contained in the current cumulative probability calculation result is determined as the number of expert networks;

[0165] The expert network corresponding to the probability weight value contained in the current cumulative probability calculation result is determined as the target expert network.

[0166] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0167] If the cumulative probability calculation result of all probability weight values ​​in the weight value sequence is less than the cumulative probability threshold, the number of expert networks in the hybrid expert model is determined as the number of expert networks, and each expert network in the hybrid expert model is determined as the target expert network.

[0168] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0169] The target expert network in the hybrid expert model analyzes and makes decisions on the word segmentation data to obtain the initial analysis and decision results;

[0170] Based on the initial analysis and decision results of each target expert network and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined.

[0171] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0172] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0173] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0174] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0175] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0176] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A data analysis method, characterized in that: The method comprises: Obtaining a text to be processed, and performing word segmentation processing on the text to be processed to obtain a word segmentation data set; For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, a weight value corresponding to each expert network in the hybrid expert model is determined; Sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results; Based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, the analysis and decision results of the word segmentation data are determined until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

2. The method according to claim 1, characterized in that For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model, and based on the analysis and processing results and the gating network, a weight value corresponding to each expert network in the hybrid expert model is determined, including: For each word segmentation data in the word segmentation data set, the word segmentation data is analyzed and processed by a hybrid expert model to obtain an analysis and processing result; Based on the analysis and processing results and the gating network in the hybrid expert model, the weight values ​​corresponding to each expert network in the hybrid expert model are calculated; the weight values ​​corresponding to the expert networks are used to characterize the decision-making efficiency of the expert networks on the word segmentation data.

3. The method according to claim 1, characterized in that The step of sorting and calculating the cumulative probability of the weight values ​​corresponding to the expert networks, and determining the number of expert networks and the target expert network based on the cumulative probability calculation results, includes: Normalizing the weight values ​​corresponding to each of the expert networks to obtain probability weight values ​​corresponding to each of the expert networks; Sorting the probability weight values ​​corresponding to the expert networks in descending order to obtain a weight value sequence; Performing cumulative probability calculation on the probability weight values ​​in the weight value sequence, and verifying the cumulative probability calculation result after each probability weight value is accumulated based on a cumulative probability threshold; If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of expert networks and the target expert network are determined.

4. The method according to claim 3, characterized in that: If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, determining the number of expert networks and the target expert network includes: If the current cumulative probability calculation result is greater than or equal to the cumulative probability threshold, the number of probability weight values ​​included in the current cumulative probability calculation result is determined as the number of expert networks; The expert network corresponding to the probability weight value included in the current cumulative probability calculation result is determined as the target expert network.

5. The method according to claim 3, characterized in that: The method further comprises: If the cumulative probability calculation result of all probability weight values ​​in the weight value sequence is less than the cumulative probability threshold, the number of expert networks in the hybrid expert model is determined as the number of expert networks, and each expert network in the hybrid expert model is determined as a target expert network.

6. The method according to claim 1, characterized in that The target expert network based on the number of expert networks and the weight value corresponding to the target expert network, determining the analysis and decision result of the word segmentation data, comprises: Analyze and make decisions on the word segmentation data through the target expert network in the hybrid expert model to obtain an initial analysis and decision result; Based on the initial analysis and decision results of each of the target expert networks and the weight values ​​corresponding to the target expert networks, the analysis and decision results of the word segmentation data are determined.

7. A data analysis device, characterized in that: The device comprises: An acquisition module is used to acquire the text to be processed and perform word segmentation processing on the text to be processed to obtain a word segmentation data set; A first determination module is used to analyze and process each word segmentation data in the word segmentation data set through a hybrid expert model, and determine a weight value corresponding to each expert network in the hybrid expert model based on the analysis and processing result and the gating network; A second determination module is used to sort the weight values ​​corresponding to the expert networks and calculate the cumulative probability, and determine the number of expert networks and the target expert network based on the cumulative probability calculation result; The third determination module is used to determine the analysis and decision results of the word segmentation data based on the target expert network of the number of expert networks and the weight value corresponding to the target expert network, until the analysis and decision results of all the word segmentation data in the word segmentation data set are obtained, and the data analysis results of the text to be processed are determined.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images