Information processing method and device, equipment, storage medium and computer program product

Abstract

The invention discloses an information processing method. The method comprises the following steps: determining a first Markov blanket of a first variable feature; wherein the first Markov blanket is a local causal network structure corresponding to the first variable feature; based on a first data feature set, performing verification management on nodes in the first Markov blanket to obtain a second Markov blanket; wherein the data features included in the first data feature set do not belong to the features included in the first Markov blanket. The invention further discloses an information processing device and equipment, a storage medium and a computer program product.

Description

Technical Field

[0001] The present application relates to the field of big data applications, and in particular to an information processing method, apparatus, device, storage medium and computer program product. Background Art

[0002] With the advancement of technology and the improvement of people's cognition, diversified data is constantly collected and stored for subsequent application analysis of the data. However, in the process of storage and application, how to automatically normalize and classify the data, dynamically build a causal graph based on the classified data, and display the causal relationship of things to users, so as to complete a construction process and enable users to have a deeper understanding of their own data has become an important research object at present. In the process of dynamically building a causal graph, the dynamic change characteristics of the entire data space can be summarized by studying the dynamic temporal causal network, that is, by constructing a directed acyclic graph. In the research process, if we are only interested in the local causal structure near a certain node, then learning the global network structure is a waste of resources and unnecessary. Therefore, local causal structure learning has received more and more attention in recent years. At present, the existing dynamic local causal structure learning mainly learns the Markov blanket set of the target node from the existing data, and then enumerates the subset of the selected parent-child nodes and spouse nodes as the condition set after the arrival of the temporal feature to judge whether the arriving temporal feature belongs to the node in the Markov blanket set, and finally outputs the obtained Markov blanket set as the local causal relationship graph of the target node.

[0003] However, after the timing feature arrives, the above implementation process uses a subset of the current parent-child node set and the spouse node set as the condition set for judgment. This is a process of enumerating a set size from an empty set to a full set, which has an exponential time complexity, resulting in low time efficiency. In addition, the reliability of the nodes in the current parent-child node set and the spouse node set is poor, resulting in poor credibility of the final output result.

[0004] Application Contents

[0005] In order to solve the above technical problems, the present application hopes to provide an information processing method, apparatus, equipment, storage medium and computer program product, which solve the problems of low time efficiency and poor credibility in the current construction of dynamic local causal networks, and propose a network construction method to improve the time efficiency of local causal network construction and ensure the credibility of the output local causal network.

[0006] The technical solution of this application is implemented as follows:

[0007] The present application provides an information processing method, the method comprising:

[0008] Determine a first Markov blanket of a first variable feature; wherein the first Markov blanket is a local causal network structure corresponding to the first variable feature;

[0009] Based on the first data feature set, verification management is performed on the nodes in the first Markov blanket to obtain a second Markov blanket; wherein the data features included in the first data feature set do not belong to the features included in the first Markov blanket.

[0010] In the above solution, based on the first data feature set, verifying and managing the nodes in the first Markov blanket to obtain the second Markov blanket includes:

[0011] Based on the first data feature set, updating a first feature set corresponding to the first Markov blanket to obtain a second feature set;

[0012] Directional processing is performed on the data features in the second feature set to obtain the second Markov blanket.

[0013] In the above solution, the updating of the first feature set corresponding to the first Markov blanket based on the first data feature set to obtain the second feature set includes:

[0014] Determining a first data feature from the first data feature set in time sequence;

[0015] Analyze the first data feature and the first feature set to obtain an analysis result;

[0016] If the analysis result shows that the first data feature belongs to the first feature set, the first data feature is added as a node to the first feature set to obtain a third feature set;

[0017] Performing independence analysis on each data feature in the first feature set and the third feature set to obtain an independence test result of each data feature in the first feature set;

[0018] Deleting data features whose independence test results are less than a first threshold in the third feature set, to obtain a fourth feature set;

[0019] The second feature set is obtained based on the first data feature set and the fourth feature set in time sequence.

[0020] In the above scheme, the method further includes:

[0021] If the analysis result indicates that the first data feature does not belong to the first feature set, obtaining a second data feature from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence;

[0022] Updating the first data feature to the second data feature;

[0023] The step of "analyzing the first data feature and the first feature set to obtain an analysis result" is performed until the second feature set is obtained.

[0024] In the above solution, obtaining the second feature set based on the first data feature set and the fourth feature set in time sequence includes:

[0025] Obtaining a second data feature from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence;

[0026] Updating the first data feature to the second data feature;

[0027] After updating the first feature set to the fourth feature set, execute the step of "if the first data feature and the first feature set are analyzed and it is determined that the first data feature belongs to the first feature set, add the first data feature as a node to the first feature set to obtain a third feature set" until all data features in the first data feature set are analyzed and processed to obtain the second feature set.

[0028] In the above solution, the independence analysis is performed on each data feature in the first feature set and the third feature set to obtain the independence test result of each data feature in the first feature set, including:

[0029] sequentially determining a fifth feature set when the third feature set does not include a data feature in the first feature set;

[0030] Based on each data feature in the first feature set and the corresponding fifth feature set that does not include the data feature, the independence test result of the corresponding data feature is determined.

[0031] In the above solution, the step of performing directional processing on the data features in the second feature set to obtain the second Markov blanket includes:

[0032] Determine a sixth feature set corresponding to the third Markov blanket for each data feature in the second feature set;

[0033] Directional processing is performed based on all the sixth feature sets corresponding to the second feature set to obtain the second Markov blanket.

[0034] In the above solution, the directional processing of all the sixth feature sets corresponding to the second feature set to obtain the second Markov blanket includes:

[0035] Taking the first variable feature as a reference, determining the connection relationship of the data features in all the sixth feature sets corresponding to the second feature set to obtain a first orientation result;

[0036] The Meek rule is used to adjust the orientation of the nodes whose first orientation results do not belong to the V structure, so as to obtain the second Markov blanket.

[0037] In the above scheme, the method further includes:

[0038] Determining the current number of depth processing for the first variable feature;

[0039] If the number of depth processing times is less than a preset number, determining a second variable feature in the second Markov blanket; wherein the second variable feature is different from the first variable feature;

[0040] obtaining a third Markov blanket of the second variable characteristics;

[0041] Updating the first Markov blanket to the third Markov blanket;

[0042] Updating the first variable feature to the second variable feature;

[0043] The first data feature set is updated to a second data feature set, and the step of "verifying and managing the nodes in the first Markov blanket to obtain a second Markov blanket" is executed until the number of deep processing times reaches the preset number of times.

[0044] In the above solution, the first Markov blanket for determining the first variable characteristic includes:

[0045] Obtain the Bayesian network corresponding to the data to be analyzed;

[0046] Based on the Bayesian network, generating a third data feature set including the first variable features;

[0047] The first Markov blanket is determined from the third data feature set using the first variable feature as a target variable.

[0048] The present application provides an information processing device, the device comprising at least: a determination unit and a management unit; wherein:

[0049] The determining unit is used to determine a first Markov blanket of the first variable feature; wherein the first Markov blanket is a local causal network structure corresponding to the first variable feature;

[0050] The management unit is used to perform verification management on the nodes in the first Markov blanket based on the first data feature set to obtain a second Markov blanket; wherein the data features included in the first data feature set do not belong to the features included in the first Markov blanket.

[0051] The present application provides an information processing device, the device comprising at least: a communication interface, a memory, a processor and a communication bus; wherein:

[0052] The memory is used to store executable instructions;

[0053] The communication bus is used to realize the communication connection between the communication interface, the processor and the memory;

[0054] The processor is used to execute the information processing program stored in the memory to implement the steps of any of the above-mentioned information processing methods.

[0055] The present application provides a storage medium, on which an information processing program is stored. When the processing program is executed, it is used to implement the steps of any of the above-mentioned information processing methods.

[0056] The present application provides a computer program product, including a computer program, which implements the steps of any of the above-mentioned information processing methods when executed by a processor.

[0057] The embodiment of the present application provides an information processing method, device, equipment, storage medium and computer program product. After determining the first Markov blanket of the first variable feature through the information processing device, the nodes in the first Markov blanket are verified and managed based on the first data feature set to obtain a second Markov blanket. In this way, the nodes in the first Markov blanket based on the first variable feature are verified and managed according to the first data feature set obtained in real time, and a more accurate second Markov blanket is obtained, so that the Markov blanket of the first variable feature is dynamically constructed, which solves the problems of low time efficiency and poor credibility when constructing a dynamic local causal network. A network construction method is proposed, which improves the time efficiency of the local causal network construction and ensures the credibility of the output local causal network. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flowchart of an information processing method provided in an embodiment of the present application;

[0059] Figure 2 Schematic diagram of an MB set provided in an embodiment of the present application Figure 1 ;

[0060] Figure 3Schematic diagram of an MB set provided in an embodiment of the present application Figure 2 ;

[0061] Figure 4 Schematic diagram of an MB set provided in an embodiment of the present application Figure 3 ;

[0062] Figure 5 Schematic diagram of an MB set provided in an embodiment of the present application Figure 4 ;

[0063] Figure 6 Schematic diagram of an MB set provided in an embodiment of the present application Figure 5 ;

[0064] Figure 7 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application;

[0065] Figure 8 A schematic diagram of the structure of an information processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0067] The embodiment of the present application provides an information processing method, referring to Figure 1 As shown, the method is applied to an information processing device, and the method comprises the following steps:

[0068] Step 101: Determine a first Markov blanket of first variable features.

[0069] Among them, the first Markov blanket is the local causal network structure corresponding to the first variable feature.

[0070] In the embodiment of the present application, the information processing device may be an electronic device with computing capabilities, such as a computer device, a server device, an intelligent mobile terminal device, etc. The first variable feature is a variable feature that the user is interested in and needs to be analyzed. The first Markov blanket of the first variable feature is a local causal network structure composed of Markov nodes related to the first variable feature.

[0071] Step 102: Based on the first data feature set, verify and manage the nodes in the first Markov blanket to obtain a second Markov blanket.

[0072] The data features included in the first data feature set do not belong to the features included in the first Markov blanket.

[0073] In an embodiment of the present application, the first data feature set is a data feature set obtained by analyzing the data collected and recorded in the application scenario related to the first variable feature that is subsequently received, and the first data feature set usually does not include the data features included in the first Markov blanket. According to the first data feature set that is subsequently received in real time, the nodes included in the first Markov blanket are verified for node reliability, and the nodes in the first Markov blanket are updated to implement verification management, thereby obtaining a second Markov blanket corresponding to the first variable feature. Among them, the depth of the second Markov blanket based on the first variable feature can be the same as the depth of the first Markov blanket, or it can be different from the depth of the first Markov blanket, which can be determined by the actual situation and is not specifically limited here.

[0074] Based on the foregoing embodiment, in other embodiments of the present application, step 101 performs verification management on the nodes in the first Markov blanket based on the first data feature set, and obtaining the second Markov blanket can be achieved by the following steps:

[0075] Based on the first data feature set, updating the first feature set corresponding to the first Markov blanket to obtain a second feature set;

[0076] Directional processing is performed on the data features in the second feature set to obtain a second Markov blanket.

[0077] In the embodiment of the present application, the first feature set corresponding to the first Markov blanket is composed of data features in the nodes included in the first Markov blanket. The first data feature set is analyzed to determine whether it belongs to the first feature set corresponding to the first Markov blanket. If there are data features belonging to the first feature set in the first data feature set, the data features belonging to the first feature set in the first data feature set are added to the first feature set to obtain the second feature set. After obtaining the second feature set, the data features in the second feature set are subjected to directional processing, that is, the causal relationship between the data features in the second feature set is determined. It should be noted that the directional processing is performed based on the first variable feature, so that the second Markov blanket of the first variable feature can be obtained.

[0078] Based on the foregoing embodiment, in other embodiments of the present application, the step of updating the first feature set corresponding to the first Markov blanket based on the first data feature set to obtain the second feature set can be implemented by the following steps:

[0079] Determining a first data feature from the first data feature set according to a time sequence;

[0080] Analyze the first data feature and the first feature set to obtain an analysis result;

[0081] If the analysis result shows that the first data feature belongs to the first feature set, the first data feature is added as a node to the first feature set to obtain a third feature set;

[0082] Performing independence analysis on each data feature in the first feature set and the third feature set to obtain an independence test result of each data feature in the first feature set;

[0083] Deleting data features whose independence test results are less than the first threshold in the third feature set to obtain a fourth feature set;

[0084] In time sequence, a second feature set is obtained based on the first data feature set and the fourth feature set.

[0085] In an embodiment of the present application, the first threshold value may be an empirical value obtained from a large number of experiments, or it may be a value determined based on the independence test results corresponding to the first variable feature and other data features in the first feature set in the current application scenario. For example, it may be the largest independence test result among the independence test results corresponding to the first variable feature and other data features in the first feature set, or it may be the average value of the corresponding independence test results, or it may be the independence test result that occurs the most times, etc.

[0086] The data features in the first data feature set can be sorted according to the received time sequence, so that the data features in the first data feature set can be obtained in sequence according to the time sequence. For example, when the first data feature is currently obtained according to the time sequence, the first data feature and the data features in the first feature set are calculated and analyzed to determine whether the first data feature belongs to the analysis result of the first feature set. When the analysis result indicates that the first data feature belongs to the first feature set, the first data feature is added to the first feature set as a feature node to obtain a third feature set; then the third feature set is used to perform independence analysis on each data feature in the first feature set and the first variable feature, so as to obtain the corresponding independence test result, and then the relationship between the independence test result of each data feature and the first threshold is compared, and the data features in the third feature set whose independence test results are less than the first threshold are deleted to obtain a fourth feature set, and the data features in the first data feature set are continuously obtained in sequence according to the time sequence, and are calculated and analyzed with the fourth feature set until the second feature set is finally obtained.

[0087] When performing independence analysis on each data feature in the first feature set and the third feature set, an independence test method may be used, such as a Pearson correlation coefficient, a chi-square test, and the like.

[0088] Based on the foregoing embodiment, in other embodiments of the present application, the method further includes the following steps:

[0089] If the analysis result shows that the first data feature does not belong to the first feature set, the second data feature is obtained from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence;

[0090] Updating the first data feature to the second data feature;

[0091] The step of “analyzing the first data feature and the first feature set to obtain an analysis result” is performed until a second feature set is obtained.

[0092] In an embodiment of the present application, when the analysis results indicate that the first data feature does not belong to the first feature set, that is, there is no correlation between the first data feature and the first variable feature, the first data feature can be discarded for the first variable feature, and the data in the first data feature set can continue to be analyzed and processed in time series. The corresponding analysis process is the same as the process of analyzing and processing the first data feature, and will not be described in detail here.

[0093] Based on the foregoing embodiment, in other embodiments of the present application, the steps are in time sequence, and the second feature set is obtained based on the first data feature set and the fourth feature set, which can also be implemented by the following steps:

[0094] Obtaining a second data feature from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence;

[0095] Updating the first data feature to the second data feature;

[0096] After updating the first feature set to the fourth feature set, execute the step of "if the first data feature and the first feature set are analyzed and it is determined that the first data feature belongs to the first feature set, add the first data feature as a node to the first feature set to obtain the third feature set" until all data features in the first data feature set are analyzed and processed to obtain the second feature set.

[0097] In an embodiment of the present application, after the analysis of the first data feature is completed, a second data feature whose time sequence is adjacent to the first data feature and follows the first data feature is obtained from the first data feature set according to the time series distribution, and is processed in the same manner as the first data feature is analyzed, and this process is repeated until all the data features in the first data feature set are analyzed to obtain a final second feature set.

[0098] Based on the foregoing embodiment, in other embodiments of the present application, the step of performing independence analysis on each data feature in the first feature set and the third feature set to obtain the independence test result of each data feature in the first feature set can be achieved by the following steps:

[0099] sequentially determining a fifth feature set when the third feature set does not include a data feature in the first feature set;

[0100] Based on each data feature in the first feature set and a corresponding fifth feature set that does not include the data feature, an independence test result of the corresponding data feature is determined.

[0101] In an embodiment of the present application, when performing independence analysis on each data feature in the first feature set and the third feature set, in order to ensure independence, the corresponding data feature in the third feature set needs to be deleted. In this way, the independence of the corresponding proposed data feature is calculated based on the fifth feature set after eliminating the corresponding data feature, and the corresponding independence test result is obtained.

[0102] Based on the foregoing embodiment, in other embodiments of the present application, the step of performing directional processing on the data features in the second feature set to obtain the second Markov blanket can be achieved by the following steps:

[0103] Determine a sixth feature set corresponding to the third Markov blanket for each data feature in the second feature set;

[0104] Directional processing is performed based on all sixth feature sets corresponding to the second feature set to obtain a second Markov blanket.

[0105] In the embodiment of the present application, the third Markov blanket of each data feature in the second feature set is the local causal network structure of each data feature in the second feature set, and the determination process is the same as the determination process of the first Markov blanket of the first variable feature, which will not be described in detail here. That is to say, when the second feature set includes N data features, N sixth feature sets can be determined to be obtained, and the N sixth feature sets are subjected to directional processing, that is, feature set fusion processing, so as to obtain the second Markov blanket corresponding to the first variable feature.

[0106] Based on the foregoing embodiment, in other embodiments of the present application, the step of performing directional processing based on all sixth feature sets corresponding to the second feature set to obtain a second Markov blanket can be achieved by the following steps:

[0107] Taking the first variable feature as a reference, determining the connection relationship of the data features in all the sixth feature sets corresponding to the second feature set, and obtaining a first orientation result;

[0108] The Meek rule is used to adjust the orientation of the nodes that do not belong to the V structure in the first orientation result to obtain the second Markov blanket.

[0109] In the embodiment of the present application, Meek's rule is: the newly oriented edge cannot generate a new V structure in the directed acyclic graph, and the newly oriented edge cannot form a ring structure with the existing directed edge.

[0110] Based on the first variable feature, the sixth feature set corresponding to all the data features included in the second feature set is judged for connection relationship, that is, the father-son relationship and the spouse relationship, to obtain a first orientation result based on the first variable feature. Because some orientations have ring structures and the like, and cannot affect the first variable feature as the benchmark, the Meek rule is used to update and adjust the edges of the first orientation result and the data features, i.e., nodes, that have not been oriented in the second feature set, so as to obtain a second Markov blanket based on the first variable feature.

[0111] Based on the foregoing embodiment, in other embodiments of the present application, the method further includes the following steps:

[0112] Determine the current number of deep processing times for the first variable feature;

[0113] If the number of deep processing times is less than the preset number, determining a second variable feature in the second Markov blanket; wherein the second variable feature is different from the first variable feature;

[0114] Obtaining a third Markov blanket of the second variable characteristics;

[0115] Update the first Markov blanket to the third Markov blanket;

[0116] Updating the first variable feature to the second variable feature;

[0117] The first data feature set is updated to the second data feature set, and the step of "verifying and managing the nodes in the first Markov blanket to obtain the second Markov blanket" is executed until the number of deep processing times reaches a preset number.

[0118] In the embodiment of the present application, the preset number of times is the preset number of depths that need to be processed for the first variable feature, which can be set according to the actual situation. When the number of depth processing is less than the preset number of times, the second variable feature is obtained from the second Markov blanket according to the preset variable storage relationship or the preset variable acquisition rule. The second variable feature is a data feature that has not been dynamically constructed based on the first variable feature. From the total Markov blanket of the first Markov blanket for obtaining the first variable feature, the third Markov blanket corresponding to the second variable feature is obtained, and then the third Markov blanket is processed in the same way as the first Markov blanket. The specific implementation can be to update the first Markov blanket to the third Markov blanket, the first variable feature to the variable feature, the first data feature set to the currently updated received second data feature set, and then perform the verification management operation of the node of the updated first Markov blanket, and repeat until the depth of the first variable feature is the preset number of times. At this time, the second Markov blanket finally obtained is determined by the second Markov blanket corresponding to the original first variable feature and the second Markov blanket corresponding to the original second variable feature, for example, it can be directional processing.

[0119] Based on the foregoing embodiment, in other embodiments of the present application, the step of determining the first Markov blanket of the first variable feature can be implemented by the following steps:

[0120] Obtain the Bayesian network corresponding to the data to be analyzed;

[0121] Based on the Bayesian network, generating a third data feature set including features of the first variable;

[0122] A first Markov blanket is determined from the third data feature set with the first variable feature as the target variable.

[0123] In an embodiment of the present application, the data to be analyzed is processed by a Bayesian processing method to obtain a Bayesian network, and then the entropy value of the data features in the Bayesian network is calculated, and then the mutual information is calculated based on the entropy value, and then an independence test calculation is performed to obtain a third data feature set including the first variable feature, and the first variable feature is used as the target variable to generate a first Markov blanket.

[0124] Based on the above embodiments, the embodiments of the present application provide a method for constructing a dynamic causal graph that is both timely and reliable, and is displayed as a dynamic causal graph. The general implementation process of the method can be: using the full set of the Markov blanket set of the target variable as the condition set to determine whether the currently arrived node is the Markov blanket of the target variable; then judging whether the newly arrived feature data belongs to the Markov blanket in the same way, if it is determined to belong to the Markov blanket, the data features included in the Markov blanket are judged in turn based on the newly added data features to see if there is a misjudgment, so as to ensure the accuracy of the method; next, the edges in the Markov blanket set are directed by judging the V structure, and if there are undirected edges, the undirected edges are processed by the Meek rule, so as to obtain the final local causal network of the target variable. If you still want to trace the target variable or view the deeper impact, that is to say, the construction of the dynamic local causal network needs to be further expanded. Repeating the above process can obtain the Markov blanket of the Markov blanket set of the target node. According to the desired depth, a set of multi-layer Markov blankets of the target variable can be constructed, so as to meet the traceability requirements. Correspondingly, the specific implementation process may include the following implementation steps:

[0125] Step a11: It is known that there is a Bayesian network BN, in which the data set corresponding to the data features of the nodes is unified with the BN probability, and the state of the BN can be represented by calculating the value of the feature probability.

[0126] Step a12: Calculate the information entropy of each node in the BN.

[0127] The information entropy of each node can be expressed as: H(X) = -∑ x P(x)logP(x), where H(X) represents the information entropy of X, X represents a random variable, P(x) represents the probability when X takes the value x, and log represents the logarithm of its subsequent value. Information entropy represents the uncertainty of the variable value, and is used to measure the influence of a certain variable on other variables and how close the connection is.

[0128] Step a13: Taking the target variable as a benchmark, calculate the mutual information between the target variable and other variables, and determine a candidate Markov Blanket (MB) set.

[0129] Among them, the mutual information between the target variable and other variables can be calculated by the following formula: In the formula, I(T;Y) represents the mutual information between T and Y, T represents the target variable, Y represents the variable currently being judged, H(T) represents the information entropy of T, H(T|Y) represents the information entropy of T under the condition of Y, P(t) represents the probability when the target variable T takes the value of t, and similarly P(t,y) represents the probability when the target variable T and the current variable Y take the values ​​of t and y at the same time. In a network of n nodes including the node of the target variable, only n-1 mutual information calculations are required, and each mutual information with the target variable is compared with the predefined δ, and the nodes with mutual information greater than δ are determined as candidate MB nodes. δ is a commonly used measurement value, which can generally be taken as 0.15.

[0130] Step a14: Use the conditional independence test to measure the quantitative independence value between each variable in the candidate MB set and the target variable.

[0131] When calculating the independence quantization value between each variable and the target variable, the independence quantization value between the excluded variable and the target variable is calculated based on the condition that the corresponding variable is not included in the candidate MB set. The conditional independence test can be performed by the formula To calculate the independence quantification value between the variable and the target variable. In the formula, G 2 Indicates the conditional independence test method currently used. Indicates that the condition set that does not include the variable currently analyzed satisfies X, where X can be the target variable, Y can be y, and Z can be z for each variable in the candidate MB set. Similarly, It indicates the number of nodes whose Y value is y and Z value is z in the conditional set excluding the variable currently analyzed. It indicates the number of nodes that satisfy the condition set of X taking the value of x and Z taking the value of z, excluding the variable currently analyzed. Indicates the number of nodes that satisfy the Z value z in the conditional set excluding the variable currently analyzed.

[0132] Step a15: According to the independence quantization value of the conditional independence test, the variables that meet the conditions are selected to form the MB set of the target variables under the current network state.

[0133] According to the independence quantization value obtained by the above conditional independence test, a node with an independence quantization value higher than the current maximum dependency is selected to obtain the MB set of the target variable. Figure 2 As shown, the variable represented by the node connected to the target variable T is the MB set of the target variable T. In this way, the MB set of the target variable is determined from the initial data set.

[0134] Step a16: Select feature Fi from the newly arrived data to obtain a feature set.

[0135] Among them, Fi represents the features arriving at time ti, the features included in the feature set are different from the features in the MB set of the target variable, i = 1, 2, ..., m.

[0136] Step a17: Based on the feature set, determine in chronological order whether the feature Fi in the feature set belongs to the MB set.

[0137] Among them, whether the feature Fi belongs to the MB set can be determined by the calculation formula in step a14. In this case, Z is Fi.

[0138] Step a18: If it is determined that the feature Fi belongs to the MB set, add the feature Fi to T's MB set.

[0139] Step a19: Taking the feature Fi added to the MB set of T excluding a certain variable as a condition set, calculate the independence test of the excluded variable to obtain the independence quantification value of the excluded variable.

[0140] The specific process may refer to the implementation process of step a14.

[0141] Step a20, determine whether the independence quantization value of the excluded variable is lower than the maximum dependency, if it is lower than the maximum dependency, delete the variable from the MB set of feature Fi added to T.

[0142] That is to say, the occurrence of T and the variable Y are independent under the condition of the set after Y is removed, that is, T⊥Y|MBT\{Y}, at this time, Y can be determined to be a false positive variable.

[0143] Step a21: Process the features in the feature set according to the process of steps a17 to a20 to construct a dynamic MB set of the target variable T.

[0144] For example, the dynamic MB set of the target variable T finally obtained can refer to Figure 3 shown.

[0145] Step a22: traverse the variables in the dynamic MB set of the target variable T and calculate the mutual information between two variables.

[0146] Step a23: Calculate the conditional mutual information between the target variable T and other variables with respect to the third variable.

[0147] The calculation formula can be written as: Among them, z is the third variable, y is the other variables, and t is the target variable.

[0148] Step a24, compare the mutual information between the target variable T and the current variable Y and the mutual information with the third variable Z as the condition set to see if there is a variable whose conditional mutual information with the third variable as the condition set is greater than the mutual information without the condition set.

[0149] Specifically, it can be determined whether I(T; Y|Z)>I(T; Y). Thus, according to the definition of the V structure, the Z node, as a conflict node (collision node), will enhance the degree of dependency between its parent nodes T and Y. Thus, according to the above judgment formula, the Z node can be determined to be a conflict node, and the three nodes T, Y and Z form a structure of T->Z<-Y. It should be noted that nodes and features are equivalent descriptions.

[0150] Step a25, mark the variable whose conditional mutual information with the third variable as the conditional set is greater than the mutual information without the conditional set as a spouse node, and modify the record in the edge orientation matrix (an N*N matrix is ​​defined to store the direction of the edge).

[0151] Among them, Figure 4 As shown, for a structure like T->B<-E, the directions of the two edges can be determined, and for other nodes in the dynamic MB set (only the parent node is left), the A->T relationship between them and the target can also be determined. In this way, the direction of the edge about the target node is determined, and whether there is an edge between the nodes in the MB set and the direction of the edge will be determined in the next step.

[0152] Step a26, traverse the dynamic MB set of the target variable T, calculate mutual information, identify the MB set corresponding to the node of the dynamic MB set, and determine the interconnection relationship between the nodes in the MB set corresponding to the node of the dynamic MB set.

[0153] Among them, whether there are nodes in the MB set connected to each other can be determined by judging whether the following relationship is satisfied That is to say, if X and Y are in each other's MB set, they can establish a direct relationship, that is, a parent-child relationship, or an indirect relationship, that is, a spouse relationship. The determination of this relationship requires the use of Meek's rule.

[0154] Step a27: further process the interconnection relationship between nodes in the MB set corresponding to the nodes based on the dynamic MB set using the Meek rule to obtain the latest Markov blanket of the target variable T.

[0155] Among them, in causal structure learning, Meek's rule is defined as follows: ① Newly directed edges cannot generate new V structures in directed acyclic graphs; ② Newly directed edges cannot form a ring structure with existing directed edges.

[0156] The specific implementation process can be recorded as follows: in the interconnected relationship of nodes in the MB set corresponding to the nodes based on the dynamic MB set, all nodes except the V structure are determined to point to T, i.e., the edge direction of X->T; then, the direct or indirect edges in the formed network are oriented by combining the V structure with the Meek rule, and the edge orientation matrix is ​​updated according to the orientation result to obtain the latest Markov blanket of the target variable T. For example, the latest Markov blanket of the target variable T obtained at this time can be as follows Figure 5 As shown, it is a dynamic local causal network with two layers, which are interconnected and have directed edges.

[0157] Furthermore, in the application process, the two-layer structure may not be sufficient to fully analyze and understand the target ratio, so a larger causal network can be constructed according to the set depth. The corresponding specific process can refer to step a28.

[0158] Step a28: After updating the dynamic MB set of the target variable T to the MB set corresponding to the latest Markov blanket according to the set depth k, repeat steps a22 to a27 k-1 times until a dynamic causal network with a depth of k corresponding to the target variable T is obtained, that is, the final Markov blanket of the target variable T.

[0159] For example, Figure 6 The figure shows a final Markov blanket of the target variable T. In the process of obtaining the final Markov blanket of the target variable T, the network structure is dynamically adjusted as new features are added, such as deleting some nodes, or adding some nodes, and giving directed edges, so that the potential mechanism of things happening can be discovered and understood.

[0160] In this way, according to the characteristics of time series data, the overall MB set is used as the condition set to judge the new MB features, which improves efficiency. In the analysis process, the data features added later are introduced as the condition set to re-test the added features, which can delete some false positive data features and greatly improve the accuracy of the method. Furthermore, k-layer construction is used for dynamic network construction, and a larger causal network is obtained according to the defined number of layers, which can simultaneously ensure the timeliness and comprehensiveness of the dynamic network structure. In this way, the Markov blanket of the target variable finally determined has an auxiliary role in understanding the laws of development of things and the potential mechanisms of events, and can be well applied in causal analysis in various fields such as medical data, genetic engineering, and bioengineering.

[0161] The information processing method provided in the embodiment of the present application determines the first Markov blanket of the first variable feature through the information processing device, and then verifies and manages the nodes in the first Markov blanket based on the first data feature set to obtain a second Markov blanket. In this way, the nodes in the first Markov blanket based on the first variable feature are verified and managed according to the first data feature set obtained in real time, and a more accurate second Markov blanket is obtained, so that the Markov blanket of the first variable feature is dynamically constructed, which solves the problems of low time efficiency and poor credibility when constructing a dynamic local causal network, and proposes a network construction method, which improves the time efficiency of the local causal network construction and ensures the credibility of the output local causal network.

[0162] Based on the above embodiments, the embodiments of the present application provide an information processing device, which can be applied to Figure 1 In the information processing method provided in the corresponding embodiment, refer to Figure 7 As shown, the information processing device 2 may include: a determination unit 21 and a management unit 22; wherein:

[0163] A determination unit 21, configured to determine a first Markov blanket of the first variable feature; wherein the first Markov blanket is a local causal network structure corresponding to the first variable feature;

[0164] The management unit 22 is used to perform verification management on the nodes in the first Markov blanket based on the first data feature set to obtain a second Markov blanket; wherein the data features included in the first data feature set do not belong to the features included in the first Markov blanket.

[0165] In other embodiments of the present application, the management unit includes: an update module and a processing module; wherein:

[0166] An updating module, configured to update a first feature set corresponding to the first Markov blanket based on the first data feature set to obtain a second feature set;

[0167] The processing module is used to perform directional processing on the data features in the second feature set to obtain a second Markov blanket.

[0168] In other embodiments of the present application, the update module is specifically used to implement the following steps:

[0169] Determining a first data feature from the first data feature set according to a time sequence;

[0170] Analyze the first data feature and the first feature set to obtain an analysis result;

[0171] If the analysis result shows that the first data feature belongs to the first feature set, the first data feature is added as a node to the first feature set to obtain a third feature set;

[0172] Performing independence analysis on each data feature in the first feature set and the third feature set to obtain an independence test result of each data feature in the first feature set;

[0173] Deleting data features whose independence test results are less than the first threshold in the third feature set to obtain a fourth feature set;

[0174] In time sequence, a second feature set is obtained based on the first data feature set and the fourth feature set.

[0175] In other embodiments of the present application, the update module is further used to implement the following steps:

[0176] If the analysis result shows that the first data feature does not belong to the first feature set, the second data feature is obtained from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence;

[0177] Updating the first data feature to the second data feature;

[0178] The step of “analyzing the first data feature and the first feature set to obtain an analysis result” is performed until a second feature set is obtained.

[0179] In other embodiments of the present application, when the update module executes the steps in a time sequence and obtains the second feature set based on the first data feature set and the fourth feature set, it can be implemented by the following steps:

[0180] Obtaining a second data feature from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence;

[0181] Updating the first data feature to the second data feature;

[0182] After updating the first feature set to the fourth feature set, execute the step of "if the first data feature and the first feature set are analyzed and it is determined that the first data feature belongs to the first feature set, add the first data feature as a node to the first feature set to obtain the third feature set" until all data features in the first data feature set are analyzed and processed to obtain the second feature set.

[0183] In other embodiments of the present application, when the updating unit performs the step of performing independence analysis on each data feature in the first feature set and the third feature set to obtain the independence test result of each data feature in the first feature set, it can be implemented by the following steps:

[0184] sequentially determining a fifth feature set when the third feature set does not include a data feature in the first feature set;

[0185] Based on each data feature in the first feature set and a corresponding fifth feature set that does not include the data feature, an independence test result of the corresponding data feature is determined.

[0186] In other embodiments of the present application, the processing module is specifically used to implement the following steps:

[0187] Determine a sixth feature set corresponding to the third Markov blanket for each data feature in the second feature set;

[0188] Directional processing is performed based on all sixth feature sets corresponding to the second feature set to obtain a second Markov blanket.

[0189] In other embodiments of the present application, when the processing module performs the directional processing based on all the sixth feature sets corresponding to the second feature set to obtain the second Markov blanket, it can be achieved by the following steps:

[0190] Taking the first variable feature as a reference, determining the connection relationship of the data features in all the sixth feature sets corresponding to the second feature set, and obtaining a first orientation result;

[0191] The Meek rule is used to adjust the orientation of the nodes that do not belong to the V structure in the first orientation result to obtain the second Markov blanket.

[0192] In other embodiments of the present application, the information processing device further includes: an acquisition unit and an update unit; wherein:

[0193] The determination unit is further used to determine the current number of depth processing times for the first variable feature;

[0194] The determination unit is further used to determine the second variable feature in the second Markov blanket if the number of depth processing times is less than a preset number; wherein the second variable feature is different from the first variable feature;

[0195] An acquisition unit, used for acquiring a third Markov blanket of the second variable feature;

[0196] An updating unit, used for updating the first Markov blanket to a third Markov blanket;

[0197] The updating unit is further used to update the first variable feature to the second variable feature;

[0198] The updating unit is further used to update the first data feature set to the second data feature set, and execute the step of "verifying and managing the nodes in the first Markov blanket to obtain the second Markov blanket" until the number of deep processing times reaches a preset number.

[0199] In other embodiments of the present application, the determination unit includes: an acquisition module, a generation module and a determination module; wherein:

[0200] An acquisition module, used to acquire the Bayesian network corresponding to the data to be analyzed;

[0201] A generating module, used for generating a third data feature set including first variable features based on a Bayesian network;

[0202] A determination module is used to determine a first Markov blanket from a third data feature set using the first variable feature as a target variable.

[0203] It should be noted that the process of information interaction between the units and modules in this embodiment can refer to the description in other embodiments and will not be repeated here.

[0204] The information processing device provided in the embodiment of the present application determines the first Markov blanket of the first variable feature through the information processing device, and then verifies and manages the nodes in the first Markov blanket based on the first data feature set to obtain a second Markov blanket. In this way, the nodes in the first Markov blanket based on the first variable feature are verified and managed according to the first data feature set obtained in real time, and a more accurate second Markov blanket is obtained, so that the Markov blanket of the first variable feature is dynamically constructed, which solves the problems of low time efficiency and poor credibility when constructing a dynamic local causal network, and proposes a network construction method, which improves the time efficiency of the local causal network construction and ensures the credibility of the output local causal network.

[0205] Based on the above embodiments, the embodiments of the present application provide an information processing device, which can be applied to Figure 1 In the information processing method provided in the corresponding embodiment, refer to Figure 8 As shown, the information processing device 3 may include: a communication interface 31, a memory 32, a processor 33 and a communication bus 34; wherein:

[0206] A memory 32, for storing executable information;

[0207] A communication bus 34, used to realize communication connection between the communication interface 31, the processor 33 and the memory 32;

[0208] The processor 33 is used to execute the information processing program stored in the memory 32 to implement the following Figure 1 The implementation process of the information processing method provided in the corresponding embodiment will not be repeated here.

[0209] Based on the foregoing embodiments, the embodiments of the present application provide a computer-readable storage medium, referred to as a storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement reference Figure 1The implementation process of the information processing method provided in the corresponding embodiment will not be repeated here.

[0210] Based on the foregoing embodiments, the embodiments of the present application further provide a computer program product, including a computer program, which can be executed by the processor 33 of the information processing device 3 to complete any of the foregoing method steps.

[0211] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0212] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0213] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0214] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0215] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. An information processing method, characterized in that: The method comprises: Determine a first Markov blanket of a first variable feature; wherein the first Markov blanket is a local causal network structure corresponding to the first variable feature; Based on the first data feature set, verification management is performed on the nodes in the first Markov blanket to obtain a second Markov blanket; wherein the data features included in the first data feature set do not belong to the features included in the first Markov blanket.

2. The method according to claim 1, characterized in that: The step of performing verification management on the nodes in the first Markov blanket based on the first data feature set to obtain a second Markov blanket includes: Based on the first data feature set, updating a first feature set corresponding to the first Markov blanket to obtain a second feature set; Directional processing is performed on the data features in the second feature set to obtain the second Markov blanket.

3. The method according to claim 2, characterized in that The updating, based on the first data feature set, of a first feature set corresponding to the first Markov blanket to obtain a second feature set includes: Determining a first data feature from the first data feature set in time sequence; Analyze the first data feature and the first feature set to obtain an analysis result; If the analysis result shows that the first data feature belongs to the first feature set, the first data feature is added as a node to the first feature set to obtain a third feature set; Performing independence analysis on each data feature in the first feature set and the third feature set to obtain an independence test result of each data feature in the first feature set; Deleting data features whose independence test results are less than a first threshold in the third feature set, to obtain a fourth feature set; The second feature set is obtained based on the first data feature set and the fourth feature set in time sequence.

4. The method according to claim 3, characterized in that The method further comprises: If the analysis result indicates that the first data feature does not belong to the first feature set, obtaining a second data feature from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence; Updating the first data feature to the second data feature; Execute the step of "analyzing the first data feature and the first feature set to obtain an analysis result" until the second feature set is obtained.

5. The method according to claim 3 or 4, characterized in that: The obtaining the second feature set based on the first data feature set and the fourth feature set in time sequence includes: Obtaining a second data feature from a subsequent data feature adjacent to the first data feature in the first data feature set in time sequence; Updating the first data feature to the second data feature; After updating the first feature set to the fourth feature set, execute the step "if the first data feature and the first feature set are analyzed and it is determined that the first data feature belongs to the first feature set, add the first data feature as a node to the first feature set to obtain a third feature set" until all data features in the first data feature set are analyzed and processed to obtain the second feature set.

6. The method according to claim 3, characterized in that The performing independence analysis on each data feature in the first feature set and the third feature set to obtain an independence test result of each data feature in the first feature set includes: sequentially determining a fifth feature set when the third feature set does not include a data feature in the first feature set; Based on each data feature in the first feature set and the corresponding fifth feature set that does not include the data feature, the independence test result of the corresponding data feature is determined.

7. The method according to claim 2, characterized in that: The step of performing directional processing on the data features in the second feature set to obtain the second Markov blanket includes: Determine a sixth feature set corresponding to the third Markov blanket for each data feature in the second feature set; Directional processing is performed based on all the sixth feature sets corresponding to the second feature set to obtain the second Markov blanket.

8. The method according to claim 7, characterized in that The performing directional processing on all the sixth feature sets corresponding to the second feature set to obtain the second Markov blanket includes: Taking the first variable feature as a reference, determining the connection relationship of the data features in all the sixth feature sets corresponding to the second feature set to obtain a first orientation result; The Meek rule is used to adjust the orientation of the nodes whose first orientation results do not belong to the V structure, so as to obtain the second Markov blanket.

9. The method according to claim 1, characterized in that: The method further comprises: Determining the current number of depth processing for the first variable feature; If the number of depth processing times is less than a preset number, determining a second variable feature in the second Markov blanket; wherein the second variable feature is different from the first variable feature; obtaining a third Markov blanket of the second variable characteristics; Updating the first Markov blanket to the third Markov blanket; Updating the first variable feature to the second variable feature; The first data feature set is updated to a second data feature set, and the step of "verifying and managing the nodes in the first Markov blanket to obtain a second Markov blanket" is executed until the number of deep processing times reaches the preset number of times.

10. The method according to claim 1, characterized in that The step of determining a first Markov blanket of first variable characteristics comprises: Obtain the Bayesian network corresponding to the data to be analyzed; Based on the Bayesian network, generating a third data feature set including the first variable features; The first Markov blanket is determined from the third data feature set using the first variable feature as a target variable.

11. An information processing device, characterized in that: The device at least comprises: a determination unit and a management unit; wherein: The determining unit is used to determine a first Markov blanket of the first variable feature; wherein the first Markov blanket is a local causal network structure corresponding to the first variable feature; The management unit is used to perform verification management on the nodes in the first Markov blanket based on the first data feature set to obtain a second Markov blanket; wherein the data features included in the first data feature set do not belong to the features included in the first Markov blanket.

12. An information processing device, characterized in that: The device at least includes: a communication interface, a memory, a processor and a communication bus; wherein: The memory is used to store executable instructions; The communication bus is used to realize the communication connection between the communication interface, the processor and the memory; The processor is used to execute the information processing program stored in the memory to implement the steps of the information processing method according to any one of claims 1 to 10.

13. A storage medium, characterized in that: The storage medium stores an information processing program, which is used to implement the steps of the information processing method according to any one of claims 1 to 10 when executed.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the steps of the information processing method according to any one of claims 1 to 10.

Images