A data processing method and device

Abstract

The present invention discloses a data processing method and device, which can establish an overall knowledge graph for each transaction to be evaluated in a transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated, and the transaction complexity of each transaction to be evaluated is determined according to the usage rate of the transaction element, the difficulty of information extraction, and the number of training corpora, and the transaction complexity of each transaction to be evaluated is sorted, and the transaction complexity corresponding to the sorting sequence number being the target sequence number is determined as the target transaction complexity, and the transaction to be evaluated corresponding to the target transaction complexity is determined as the transaction to be learned by the network point robot. The present invention can effectively determine the transaction to be learned by the network point robot.

Description

Technical Field

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

[0002] With the advancement of intelligent science and technology, banks are continuously expanding the deployment and application of branch robots.

[0003] Branch robots can be transaction processing devices deployed at bank branches. They can learn transaction processing procedures, handle transaction requests raised by customers, and solve customer needs based on chatting with customers and helping to answer their questions. Among them, bank branches need to process many types of transactions, and branch robots cannot fully cover all transactions due to limited processing resources.

[0004] However, existing technologies cannot effectively determine transactions to be learned by branch robots. Summary of the invention

[0005] In view of the above problems, the present invention provides a data processing method and device that overcomes the above problems or at least partially solves the above problems. The technical solution is as follows:

[0006] A data processing method, comprising:

[0007] For each transaction to be evaluated in the transaction set to be evaluated, an overall knowledge graph is established, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated;

[0008] Determine the transaction complexity of each transaction to be evaluated according to the usage rate of transaction elements, the difficulty of information extraction and the amount of training corpus;

[0009] Sorting the transaction complexity of each of the transactions to be evaluated, and determining the transaction complexity corresponding to the target sequence number as the target transaction complexity;

[0010] The transaction to be evaluated corresponding to the target transaction complexity is determined as the transaction to be learned by the branch robot.

[0011] Optionally, for each transaction to be evaluated in the transaction set to be evaluated, establishing an overall knowledge graph includes:

[0012] Establishing knowledge graphs for the transactions to be evaluated respectively;

[0013] Based on the upstream and downstream relationships between the transactions to be evaluated, the overall knowledge graph is established using the knowledge graphs of the transactions to be evaluated.

[0014] Optionally, the sorting of the transaction complexity of each of the transactions to be evaluated includes:

[0015] Sorting the transaction complexity of each transaction to be evaluated from low to high;

[0016] The step of determining the transaction complexity corresponding to the target sequence number as the target transaction complexity includes:

[0017] The transaction complexity corresponding to the sorting sequence number that is not greater than the preset sequence number threshold is determined as the target transaction complexity.

[0018] Optionally, determining the transaction complexity of each transaction to be evaluated according to the usage rate of transaction elements, the difficulty of information extraction and the amount of training corpus, includes:

[0019] Determine one of the transactions to be evaluated in the set of transactions to be evaluated as a current target transaction to be evaluated;

[0020] Determine the complexity of each transaction element in the current target transaction to be evaluated based on the usage rate of the transaction element, the difficulty of information extraction, and the amount of training corpus;

[0021] The average of the complexity of each transaction element in the current target transaction to be evaluated is determined as the transaction complexity of the current target transaction to be evaluated;

[0022] Return to the step of determining one of the transactions to be evaluated in the transaction set to be evaluated as the current target transaction to be evaluated, until the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated is determined.

[0023] Optionally, determining the complexity of each transaction element in the current target transaction to be evaluated according to the usage rate of the transaction element, the difficulty of information extraction and the amount of training corpus, includes:

[0024] Determine a transaction element in the current target transaction to be evaluated as the current target transaction element;

[0025] Determine the current utilization rate of target transaction elements;

[0026] Determine the difficulty of extracting information about the current target transaction elements;

[0027] Determine the amount of training corpus for the current target transaction element;

[0028] Determine the complexity of the current target transaction elements based on the usage rate of the current target transaction elements, the difficulty of information extraction, and the amount of training corpus;

[0029] Return to the step of determining a transaction element in the current target transaction to be evaluated as the current target transaction element, until the complexity of each transaction element in the current target transaction to be evaluated is determined.

[0030] Optionally, determining the current usage rate of the target transaction element includes:

[0031] The sum of the transaction volume proportions of the first transactions to be evaluated is determined as the usage rate of the current target transaction element, wherein the first transactions to be evaluated are the transactions to be evaluated that include the current target transaction element in the transaction set to be evaluated.

[0032] Optionally, determining the difficulty of extracting information of the current target transaction element includes:

[0033] Calculating the cosine distance between the current target transaction element and other transaction elements in the overall knowledge graph;

[0034] The calculated cosine distances are sorted in descending order, and the cosine distances whose sorting numbers are not greater than a preset threshold are determined as target cosine distances;

[0035] Calculating the average of the target cosine distances;

[0036] The calculated reciprocal value of the average value is determined as the information extraction difficulty of the current target transaction element.

[0037] Optionally, determining the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction, and the amount of training corpus includes:

[0038] Normalize the usage rate of the current target transaction element to obtain the processed usage rate;

[0039] Normalizing the difficulty of information extraction of the current target transaction element to obtain the difficulty of information extraction after processing;

[0040] Normalize the number of training corpora of the current target transaction element to obtain the number of training corpora after processing;

[0041] The post-processing utilization rate, the post-processing information extraction difficulty and the post-processing training corpus quantity are input into the complexity calculation formula:

[0042] S=r+n-df

[0043] In which, S is the complexity of the current target transaction element, r is the utilization rate of the current target transaction element after processing, n is the difficulty of information extraction after processing of the current target transaction element, and df is the number of training corpora after processing of the current target transaction element;

[0044] The complexity of the current target transaction element output by the complexity calculation formula is obtained.

[0045] A data processing device comprises: a first establishing unit, a first determining unit, a first sorting unit, a second determining unit and a third determining unit, wherein:

[0046] The first establishing unit is configured to execute: establishing an overall knowledge graph for each transaction to be evaluated in the transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated;

[0047] The first determination unit is configured to execute: determining the transaction complexity of each of the transactions to be evaluated according to the usage rate of transaction elements, the difficulty of information extraction and the amount of training corpus;

[0048] The first sorting unit is configured to execute: sorting the transaction complexity of each of the transactions to be evaluated;

[0049] The second determining unit is configured to execute: determining the transaction complexity corresponding to the target sequence number as the target transaction complexity;

[0050] The third determination unit is configured to execute: determining the transaction to be evaluated corresponding to the target transaction complexity as the transaction to be learned by the branch robot.

[0051] Optionally, the first establishing unit includes: a second establishing unit and a third establishing unit, wherein:

[0052] The second establishing unit is configured to execute: establishing a knowledge graph of each of the transactions to be evaluated respectively;

[0053] The third establishing unit is configured to execute: based on the upstream and downstream relationships between the transactions to be evaluated, using the knowledge graphs of the transactions to be evaluated to establish the overall knowledge graph.

[0054] Optionally, the first sorting unit is configured to execute: sorting the transaction complexity of each of the transactions to be evaluated in order from low to high;

[0055] The second determining unit is configured to execute: determining the transaction complexity corresponding to the sorting sequence number that is not greater than a preset sequence number threshold as the target transaction complexity.

[0056] Optionally, the first determining unit includes: a fourth determining unit, a fifth determining unit, a sixth determining unit and a first triggering unit, wherein:

[0057] The fourth determining unit is configured to execute: determining one of the transactions to be evaluated in the set of transactions to be evaluated as a current target transaction to be evaluated;

[0058] The fifth determining unit is configured to execute: determining the complexity of each transaction element in the current target transaction to be evaluated according to the usage rate of the transaction element, the difficulty of information extraction and the amount of training corpus;

[0059] The sixth determining unit is configured to execute: determining the average of the complexity of each transaction element in the current target transaction to be evaluated as the transaction complexity of the current target transaction to be evaluated;

[0060] The first triggering unit is configured to execute: triggering the fourth determining unit until the transaction complexity of each transaction to be evaluated in the set of transactions to be evaluated is determined.

[0061] Optionally, the fifth determination unit includes: a seventh determination unit, a usage rate determination unit, a difficulty determination unit, a corpus quantity determination unit, a complexity determination unit, and a second triggering unit, wherein:

[0062] The seventh determining unit is configured to execute: determining a transaction element in the current target transaction to be evaluated as the current target transaction element;

[0063] The usage rate determination unit is configured to perform: determining the usage rate of the current target transaction element;

[0064] The difficulty determination unit is configured to perform: determining the difficulty of information extraction of the current target transaction element;

[0065] The corpus quantity determination unit is configured to perform: determining the training corpus quantity of the current target transaction element;

[0066] The complexity determination unit is configured to perform: determining the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction and the amount of training corpus;

[0067] The second triggering unit is configured to execute: triggering the seventh determining unit until the complexity of each transaction element in the current target transaction to be evaluated is determined.

[0068] Optionally, the usage rate determination unit includes: determining the sum of the transaction volume proportions of each first transaction to be evaluated as the usage rate of the current target transaction element, and the first transaction to be evaluated is the transaction to be evaluated that includes the current target transaction element in the transaction set to be evaluated.

[0069] Optionally, the difficulty determination unit includes: a cosine distance calculation unit, a second sorting unit, an eighth determination unit, an average value calculation unit and a ninth determination unit, wherein:

[0070] The cosine distance calculation unit is configured to perform: calculating the cosine distances between the current target transaction element and other transaction elements in the overall knowledge graph;

[0071] The second sorting unit is configured to execute: sorting the calculated cosine distances in descending order;

[0072] The eighth determination unit is configured to execute: determining all cosine distances whose sorting numbers are not greater than a preset threshold as target cosine distances;

[0073] The average value calculation unit is configured to perform: calculating the average value of each of the target cosine distances;

[0074] The ninth determining unit is configured to execute: determining the calculated reciprocal value of the average value as the information extraction difficulty of the current target transaction element.

[0075] Optionally, the complexity determination unit includes: a first processing unit, a first obtaining unit, a second processing unit, a second obtaining unit, a third processing unit, a third obtaining unit, an input unit and a fourth obtaining unit, wherein:

[0076] The first processing unit is configured to perform: normalizing the usage rate of the current target transaction element;

[0077] The first obtaining unit is configured to execute: obtaining a post-processing usage rate;

[0078] The second processing unit is configured to perform: normalizing the difficulty of information extraction of the current target transaction element;

[0079] The second obtaining unit is configured to perform: obtaining a difficulty of extracting processed information;

[0080] The third processing unit is configured to perform: normalizing the number of training corpora of the current target transaction element;

[0081] The third obtaining unit is configured to execute: obtaining the amount of processed training corpus;

[0082] The input unit is configured to execute: inputting the post-processing utilization rate, the post-processing information extraction difficulty and the post-processing training corpus quantity into the complexity calculation formula:

[0083] S=r+n-df

[0084] In which, S is the complexity of the current target transaction element, r is the utilization rate of the current target transaction element after processing, n is the difficulty of information extraction after processing of the current target transaction element, and df is the number of training corpora after processing of the current target transaction element;

[0085] The fourth obtaining unit is configured to execute: obtaining the complexity of the current target transaction element output by the complexity calculation formula.

[0086] The data processing method and device proposed in this embodiment can establish an overall knowledge graph for each transaction to be evaluated in the transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated. According to the usage rate of the transaction elements, the difficulty of information extraction, and the number of training corpora, the transaction complexity of each transaction to be evaluated is determined respectively, and the transaction complexity of each transaction to be evaluated is sorted, and the transaction complexity corresponding to the sorting sequence number being the target sequence number is determined as the target transaction complexity, and the transaction to be evaluated corresponding to the target transaction complexity is determined as the transaction to be learned by the branch robot. The present invention can effectively determine the transaction to be learned by the branch robot.

[0087] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0089] Figure 1 A flow chart of a first data processing method provided by an embodiment of the present invention is shown;

[0090] Figure 2 A knowledge graph of a loss reporting transaction provided by an embodiment of the present invention is shown;

[0091] Figure 3 An overall knowledge graph provided by an embodiment of the present invention is shown;

[0092] Figure 4 A flow chart of a second data processing method provided by an embodiment of the present invention is shown;

[0093] Figure 5 A schematic diagram showing the structure of a first data processing device provided by an embodiment of the present invention is shown;

[0094] Figure 6 A schematic structural diagram of a second data processing device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0095] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0096] like Figure 1 As shown, this embodiment proposes a first data processing method. The method may include the following steps:

[0097] S101. For each transaction to be evaluated in the transaction set to be evaluated, establish an overall knowledge graph, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated;

[0098] It should be noted that the present invention can be applied to network point robots, and can also be applied to electronic devices such as mobile phones and tablet computers.

[0099] The transactions to be evaluated in the transaction set to be evaluated may be formulated by the technical staff of the bank according to the actual transaction processing situation of the bank outlets, and the present invention does not limit this.

[0100] The overall knowledge graph may be composed of the knowledge graphs of each transaction to be evaluated.

[0101] Optionally, in other data processing methods proposed in this embodiment, step S101 may include:

[0102] Establish a knowledge graph for each transaction to be evaluated;

[0103] Based on the upstream and downstream relationships between the transactions to be evaluated, the overall knowledge graph is established using the knowledge graphs of the transactions to be evaluated.

[0104] The knowledge graph of the transaction to be evaluated may include the transaction to be evaluated and its transaction elements. For example, the knowledge graph of “reporting loss” may include: Figure 2 As shown in the knowledge graph of "reporting loss", the card number and password are the transaction elements of "reporting loss".

[0105] Specifically, the present invention can construct an overall knowledge graph using the knowledge graph of each transaction to be evaluated based on the upstream and downstream relationships between each transaction to be evaluated. Figure 3 In the overall knowledge graph shown, "card replacement" is a downstream transaction of "reporting loss", that is, before applying for "card replacement", you need to apply for "card replacement" first, and the transaction elements of "card replacement" include a password.

[0106] It should be noted that transactions to be evaluated that do not have upstream or downstream relationships with other transactions to be evaluated can also be included in the overall knowledge graph. For example, Figure 3 The transaction elements of "transfer" in the overall knowledge graph shown include the payee, amount and receiving account number. There is no upstream or downstream relationship between "transfer" and other transactions to be evaluated in the overall knowledge graph.

[0107] In the process of constructing the overall knowledge graph, the present invention can disambiguate the transaction elements in each transaction to be evaluated, that is, the transaction elements with the same meaning but different names can be uniformly described using the same name. For example, for "account number" and "account", the present invention can disambiguate the two transaction elements and uniformly use the name "account number" to describe them.

[0108] S102, determining the transaction complexity of each transaction to be evaluated according to the usage rate of transaction elements, the difficulty of information extraction and the amount of training corpus;

[0109] The usage rate of the transaction element may be the usage rate of the transaction element in each transaction to be evaluated. Specifically, the usage rate of the transaction element may be the sum of the transaction volume proportions of each transaction to be evaluated that includes the transaction element. For example, the usage rate of the first transaction element may be the sum of the transaction volume proportions of each transaction to be evaluated that includes the first transaction element.

[0110] It should be noted that the present invention can collect text sentences or voice sentences output by customers during the process of actually handling transactions requested by customers at bank branches, and manually annotate the transaction elements present in the sentences, so as to obtain training corpus that can be used to train the information extraction model, and use the training corpus to train the information extraction model to obtain a trained information extraction model, and use the information extraction model to extract transaction elements from the sentences input by customers.

[0111] The information extraction difficulty of the transaction element may be the difficulty of the information extraction model identifying and extracting the transaction element from a sentence containing the transaction element when the customer inputs the sentence. For example, the information extraction difficulty of the first transaction element may be the difficulty of the information extraction model identifying and extracting the first transaction element from the sentence when the customer inputs the sentence containing the first transaction element.

[0112] The number of training corpora of transaction elements may be the number of training corpora containing transaction elements.

[0113] Specifically, the present invention can determine the complexity of the transaction element according to the usage rate, information extraction difficulty and the number of training corpus of the transaction element. For example, for the first transaction element, the present invention can determine the complexity of the first transaction element according to the usage rate, information extraction difficulty and the number of training corpus of the first transaction element; for another example, for the second transaction element, the present invention can determine the complexity of the second transaction element according to the usage rate, information extraction difficulty and the number of training corpus of the second transaction element.

[0114] It is understandable that the present invention can determine the transaction complexity of a transaction to be evaluated based on the complexity of each transaction element in a transaction to be evaluated. For example, for a first transaction to be evaluated, the present invention can determine the transaction complexity of the first transaction to be evaluated based on the complexity of each transaction element in the first transaction to be evaluated; for another example, for a second transaction to be evaluated, the present invention can determine the transaction complexity of the second transaction to be evaluated based on the complexity of each transaction element in the second transaction to be evaluated.

[0115] S103, sorting the transaction complexity of each transaction to be evaluated;

[0116] Specifically, the present invention can sort the transaction complexity of each transaction to be evaluated after determining the transaction complexity of each transaction to be evaluated.

[0117] Optionally, the present invention may sort the complexity of each transaction in order from high to low; the present invention may also sort the complexity of each transaction in order from low to high.

[0118] S104, determining the transaction complexity corresponding to the target sequence number as the target transaction complexity;

[0119] The target sequence number may be a sequence number corresponding to a high transaction complexity or a low transaction complexity. The target sequence number may include one or more sorting sequence numbers, which is not limited in the present invention.

[0120] It should be noted that the present invention can determine high transaction complexity from the sorted transaction complexities, and select the transactions to be evaluated corresponding to the high transaction complexity as the transactions to be learned by the branch robots (at this time, the high transaction complexity is the target transaction complexity), so that the branch robots can handle transactions with high transaction complexity, while improving transaction processing efficiency and reducing the human resources consumed by manual processing of complex transactions.

[0121] Optionally, the present invention can also determine low transaction complexity from the sorted transaction complexities, and select the transactions to be evaluated corresponding to the low transaction complexity as the transactions to be learned by the branch robots (in this case, the low transaction complexity is the target transaction complexity), so that the branch robots can handle transactions with low transaction complexity, while improving the transaction handling efficiency, reducing the learning difficulty of the branch robots for transaction handling, and ensuring the success rate of the branch robots in handling transactions.

[0122] Optionally, when sorting the transaction complexities in descending order, the present invention may determine a certain number of transaction complexities sorted in the front column as high transaction complexity, and determine a certain number of transaction complexities sorted in the back column as low transaction complexity. At this time, if the transactions to be evaluated corresponding to the high transaction complexity are determined as the transactions to be learned by the branch robot, the target serial number may include a serial number not greater than a preset threshold; if the transactions to be evaluated corresponding to the low transaction complexity are determined as the transactions to be learned by the branch robot, the target serial number may include a serial number not less than a preset threshold.

[0123] It is understandable that, after determining the target serial number, the present invention can determine the transaction complexity corresponding to the target serial number as the target transaction complexity.

[0124] S105. Determine the transaction to be evaluated corresponding to the target transaction complexity as the transaction to be learned by the branch robot.

[0125] Specifically, after determining the target transaction complexity, the present invention can determine the transaction to be evaluated corresponding to the target transaction complexity as the transaction to be learned by the branch robot.

[0126] Optionally, in other data processing methods proposed in this embodiment, step S104 can be specifically as follows: sorting the transaction complexity of each transaction to be evaluated in order from low to high; step S105 can be specifically as follows: determining the transaction complexity corresponding to the sorting number that is not greater than a preset sequence number threshold as the target transaction complexity.

[0127] Among them, the present invention can sort the transaction complexity from low to high, and determine a certain number of sorting numbers in the front row as the target numbers. For example, the present invention can determine the top 5 sorting numbers as the target numbers. At this time, the present invention can determine the transactions to be evaluated corresponding to the low transaction complexity as the transactions to be learned by the branch robot.

[0128] The preset sequence number threshold may be formulated by technical personnel according to actual working conditions, and the present invention does not limit this.

[0129] The data processing method proposed in this embodiment can establish an overall knowledge graph for each transaction to be evaluated in the transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated. According to the usage rate of the transaction elements, the difficulty of information extraction, and the number of training corpora, the transaction complexity of each transaction to be evaluated is determined respectively, and the transaction complexity of each transaction to be evaluated is sorted, and the transaction complexity corresponding to the sorting sequence number being the target sequence number is determined as the target transaction complexity, and the transaction to be evaluated corresponding to the target transaction complexity is determined as the transaction to be learned by the network point robot. The present invention can effectively determine the transaction to be learned by the network point robot.

[0130] based on Figure 1 The steps shown are as follows: Figure 4 As shown, this embodiment proposes a second data processing method. In this method, step S102 may include the following steps:

[0131] S201, determining a transaction to be evaluated in the set of transactions to be evaluated as a current target transaction to be evaluated;

[0132] It should be noted that the present invention can obtain the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated by successively determining the transactions to be evaluated in the transaction set to be evaluated as the current target transactions to be evaluated and determining the transaction complexity of the current target transactions to be evaluated.

[0133] S202, determining the complexity of each transaction element in the current target transaction to be evaluated according to the usage rate of the transaction element, the difficulty of information extraction and the amount of training corpus;

[0134] Specifically, the present invention can determine the complexity of a transaction element according to the usage rate, information extraction difficulty, and the amount of training corpus of a transaction element. For example, for the first transaction element in the current target transaction to be evaluated, the present invention can determine the complexity of the first transaction element according to the usage rate, information extraction difficulty, and the amount of training corpus of the first transaction element.

[0135] S203, determining the average of the complexity of each transaction element in the current target transaction to be evaluated as the transaction complexity of the current target transaction to be evaluated;

[0136] Specifically, after determining the complexity of each transaction element in the current target transaction to be evaluated, the present invention can calculate the average of the complexity of each transaction element, and determine the calculated average as the transaction complexity of the current target transaction to be evaluated.

[0137] S204 , returning to step S201 , until the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated is determined.

[0138] It can be understood that the present invention can determine the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated by executing steps S201, S202, S203 and S204.

[0139] The data processing method proposed in this embodiment can obtain the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated by successively determining the transactions to be evaluated in the transaction set to be evaluated as the current target transactions to be evaluated and determining the transaction complexity of the current target transactions to be evaluated, thereby ensuring the success rate of obtaining the transaction complexity of each transaction to be evaluated.

[0140] based on Figure 4 The steps shown in the figure, this embodiment proposes a third data processing method. In this method, step S202 may include the following steps:

[0141] S301, determining a transaction element in the current target transaction to be evaluated as the current target transaction element;

[0142] The present invention can obtain the complexity of each transaction element in the current target transaction to be evaluated by sequentially determining each transaction element in the current target transaction to be evaluated as the current target transaction element and calculating the complexity of the current target transaction element.

[0143] S302, determining the current utilization rate of the target transaction element;

[0144] Optionally, step S302 may include:

[0145] The sum of the transaction volume proportions of the first transactions to be evaluated is determined as the usage rate of the current target transaction element. The first transaction to be evaluated is the transaction to be evaluated that includes the current target transaction element in the transaction set to be evaluated.

[0146] The transaction volume ratio of a certain transaction may be a ratio of the transaction volume of the transaction to the transaction volumes of all transactions to be evaluated in the transaction set to be evaluated.

[0147] Specifically, the present invention can determine the sum of the transaction volume ratios of each transaction to be evaluated that includes the current target transaction element as the utilization rate of the current target transaction element. For example, if there are two first transactions to be evaluated that include the current target transaction element, and their transaction volume ratios are 0.2 and 0.5 respectively, the utilization rate of the current target transaction element can be 0.7.

[0148] S303, determining the difficulty of extracting information of the current target transaction element;

[0149] Optionally, step S303 may include:

[0150] Calculate the cosine distance between the current target transaction element and other transaction elements in the overall knowledge graph;

[0151] The calculated cosine distances are sorted in descending order, and the cosine distances whose sorting numbers are not greater than a preset threshold are determined as target cosine distances;

[0152] Calculate the average cosine distance of each target;

[0153] The reciprocal value of the calculated average value is determined as the information extraction difficulty of the current target transaction element.

[0154] It is understandable that when the degree of distinction between a transaction element and other transaction elements is greater, the difficulty of extracting its information is smaller; when the degree of distinction between a transaction element and other transaction elements is smaller, the difficulty of extracting its information is greater. Among them, the present invention can determine the degree of distinction between a transaction element and other transaction elements based on the cosine distance between a transaction element and other transaction elements.

[0155] Among them, in the process of calculating the cosine distance between transaction elements, the present invention can first use the word vector model to perform text vectorization on each transaction element in the overall knowledge graph to obtain each transaction element represented by text vectorization. Then, for the two transaction elements whose cosine distance is to be calculated, the present invention calculates the cosine distance of the text vectors corresponding to the two transaction elements according to the cosine distance calculation formula between vectors to obtain the cosine distance between the two transaction elements.

[0156] Specifically, after calculating the cosine distance between the current target transaction element and other transaction elements in the overall knowledge graph, the cosine distances can be sorted in descending order, and a certain number of cosine distances in the front row can be determined as the target cosine distances. For example, the first 10% of the cosine distances can be determined as the target cosine distances. Afterwards, the present invention can determine the average value of the target cosine distance as the discrimination between the current target transaction element and other transaction elements, and determine the reciprocal of the discrimination as the information extraction difficulty of the current target transaction element.

[0157] S304, determining the amount of training corpus of the current target transaction element;

[0158] Specifically, the present invention may determine the amount of training corpus for the current target transaction element used by the information extraction model in the training phase as the amount of training corpus for the current target transaction element.

[0159] S305, determining the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction, and the amount of training corpus;

[0160] Specifically, the present invention can use a certain calculation formula to calculate the complexity of the current target transaction element according to the current usage rate of the target transaction element, the difficulty of information extraction and the number of training corpora. For example, the usage rate of the current target transaction element can be added to the number of training corpora to obtain the corresponding sum, and then the difference obtained by subtracting the information extraction difficulty from the sum is determined as the complexity of the current target transaction element; for example, corresponding weights can be assigned to the usage rate, the difficulty of information extraction and the number of training corpora, and then the above addition and subtraction calculation process is performed in combination with the weights.

[0161] S306: Return to step S301 until the complexity of each transaction element in the current target transaction to be evaluated is determined.

[0162] It is understandable that the present invention may execute steps S301, S302, S303, S304, S305 and S306 in a loop until the complexity of each transaction element in the current target transaction to be evaluated is determined.

[0163] The data processing method proposed in this embodiment can obtain the complexity of each transaction element in the current target transaction to be evaluated by determining each transaction element in the current target transaction to be evaluated as the current target transaction element in turn and calculating the complexity of the current target transaction element, thereby ensuring the success rate of obtaining the complexity of each transaction element in the current target transaction to be evaluated, thereby further ensuring the success rate of obtaining the transaction complexity of each transaction to be evaluated.

[0164] Based on the third data processing method described above, this embodiment proposes a fourth data processing method. In the fourth data processing method, step S305 may include the following steps:

[0165] Normalize the usage rate of the current target transaction element to obtain the processed usage rate;

[0166] Normalizing the difficulty of information extraction of the current target transaction element to obtain the difficulty of information extraction after processing;

[0167] Normalize the number of training corpora of the current target transaction element to obtain the number of training corpora after processing;

[0168] Input the post-processing usage rate, post-processing information extraction difficulty, and post-processing training corpus quantity into the complexity calculation formula:

[0169] S=r+n-df

[0170] In which, S is the complexity of the current target transaction element, r is the utilization rate of the current target transaction element after processing, n is the difficulty of information extraction after processing of the current target transaction element, and df is the number of training corpora after processing of the current target transaction element;

[0171] Obtain the complexity of the current target transaction element output by the complexity calculation formula.

[0172] It can be understood that in the process of calculating complexity based on usage rate, information extraction difficulty and the number of training corpora, in order to effectively improve the accuracy of quantitative measurement of complexity, the present invention can first apply normalization processing technology to normalize the usage rate, information extraction difficulty and the number of training corpora respectively, and then use the normalized usage rate, information extraction difficulty and the number of training corpora to calculate the complexity.

[0173] Among them, after obtaining the usage rate, information extraction difficulty and number of training corpora of each transaction element in the overall knowledge graph, the present invention can perform normalization processing on the usage rate, information extraction difficulty and number of training corpora of each transaction element respectively, and map the values ​​of the usage rate, information extraction difficulty and number of training corpora to the interval range of 0 to 1. For example, when the present invention performs normalization processing on the usage rate of the current target transaction element, it can perform normalization processing according to the usage rate of the current target transaction element and the usage rate of each transaction element in the overall knowledge graph.

[0174] Specifically, the present invention can calculate the complexity of the current target transaction element according to the complexity calculation formula after obtaining the post-processing utilization rate, post-processing information extraction difficulty and post-processing training corpus quantity of the current target transaction element.

[0175] The data processing method proposed in this embodiment can first apply normalization processing technology to normalize the usage rate, information extraction difficulty and the number of training corpora respectively, and then use the normalized usage rate, information extraction difficulty and training corpus to calculate the complexity, thereby effectively improving the accuracy of quantitative calculation of the complexity of transaction elements.

[0176] and Figure 1 The method shown corresponds to Figure 5 As shown, this embodiment proposes a first data processing device. The device may include: a first establishing unit 101, a first determining unit 102, a first sorting unit 103, a second determining unit 104 and a third determining unit 105, wherein:

[0177] The first establishing unit 101 is configured to execute: establishing an overall knowledge graph for each transaction to be evaluated in the transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated;

[0178] It should be noted that the present invention can be applied to network point robots, and can also be applied to electronic devices such as mobile phones and tablet computers.

[0179] The transactions to be evaluated in the transaction set to be evaluated may be formulated by the technical staff of the bank according to the actual transaction processing situation of the bank outlets, and the present invention does not limit this.

[0180] The overall knowledge graph may be composed of the knowledge graphs of each transaction to be evaluated.

[0181] Optionally, in other data processing devices provided in this embodiment, the first establishing unit 101 may include: a second establishing unit and a third establishing unit, wherein:

[0182] The second establishing unit is configured to execute: establishing a knowledge graph for each transaction to be evaluated;

[0183] The third establishing unit is configured to execute: based on the upstream and downstream relationships between the transactions to be evaluated, using the knowledge graphs of the transactions to be evaluated to establish an overall knowledge graph.

[0184] Among them, the knowledge graph of the transaction to be evaluated may include the transaction to be evaluated and its transaction elements.

[0185] Specifically, the present invention can construct an overall knowledge graph using the knowledge graph of each transaction to be evaluated according to the upstream and downstream relationships between the transactions to be evaluated. It should be noted that transactions to be evaluated that do not have an upstream and downstream relationship with other transactions to be evaluated can also be included in the overall knowledge graph.

[0186] In the process of constructing the overall knowledge graph, the present invention can disambiguate the transaction elements in each transaction to be evaluated.

[0187] The first determining unit 102 is configured to execute: determining the transaction complexity of each transaction to be evaluated according to the usage rate of the transaction elements, the difficulty of information extraction and the amount of training corpus;

[0188] The usage rate of the transaction element may be the usage rate of the transaction element in each transaction to be evaluated. Specifically, the usage rate of the transaction element may be the sum of the transaction volume proportions of each transaction to be evaluated that contains the transaction element.

[0189] It should be noted that the present invention can collect text sentences or voice sentences output by customers during the process of actually handling transactions requested by customers at bank branches, and manually annotate the transaction elements present in the sentences, so as to obtain training corpus that can be used to train the information extraction model, and use the training corpus to train the information extraction model to obtain a trained information extraction model, and use the information extraction model to extract transaction elements from the sentences input by customers.

[0190] The difficulty of extracting information about transaction elements may be the difficulty of an information extraction model identifying and extracting transaction elements from a sentence containing transaction elements when a customer inputs the sentence.

[0191] The number of training corpora of transaction elements may be the number of training corpora containing transaction elements.

[0192] Specifically, the present invention can determine the complexity of transaction elements according to the usage rate of transaction elements, the difficulty of information extraction and the amount of training corpus.

[0193] It is understandable that the present invention can determine the transaction complexity of a transaction to be evaluated according to the complexity of each transaction element in the transaction to be evaluated.

[0194] The first sorting unit 103 is configured to perform: sorting the transaction complexity of each transaction to be evaluated;

[0195] Specifically, the present invention can sort the transaction complexity of each transaction to be evaluated after determining the transaction complexity of each transaction to be evaluated.

[0196] Optionally, the present invention may sort the complexity of each transaction in order from high to low; the present invention may also sort the complexity of each transaction in order from low to high.

[0197] The second determining unit 104 is configured to execute: determining the transaction complexity corresponding to the sorting sequence number being the target sequence number as the target transaction complexity;

[0198] The target sequence number may be a sequence number corresponding to a high transaction complexity or a low transaction complexity. The target sequence number may include one or more sorting sequence numbers, which is not limited in the present invention.

[0199] It should be noted that the present invention can determine high transaction complexity from the sorted transaction complexities, and select the transactions to be evaluated corresponding to the high transaction complexity as the transactions to be learned by the branch robots (at this time, the high transaction complexity is the target transaction complexity), so that the branch robots can handle transactions with high transaction complexity, while improving transaction processing efficiency and reducing the human resources consumed by manual processing of complex transactions.

[0200] Optionally, the present invention can also determine low transaction complexity from the sorted transaction complexities, and select the transactions to be evaluated corresponding to the low transaction complexity as the transactions to be learned by the branch robots (in this case, the low transaction complexity is the target transaction complexity), so that the branch robots can handle transactions with low transaction complexity, while improving the transaction handling efficiency, reducing the learning difficulty of the branch robots for transaction handling, and ensuring the success rate of the branch robots in handling transactions.

[0201] Optionally, when sorting the transaction complexities in descending order, the present invention may determine a certain number of transaction complexities sorted in the front column as high transaction complexity, and determine a certain number of transaction complexities sorted in the back column as low transaction complexity. At this time, if the transactions to be evaluated corresponding to the high transaction complexity are determined as the transactions to be learned by the branch robot, the target serial number may include a serial number not greater than a preset threshold; if the transactions to be evaluated corresponding to the low transaction complexity are determined as the transactions to be learned by the branch robot, the target serial number may include a serial number not less than a preset threshold.

[0202] It is understandable that, after determining the target serial number, the present invention can determine the transaction complexity corresponding to the target serial number as the target transaction complexity.

[0203] The third determining unit 105 is configured to execute: determining the transaction to be evaluated corresponding to the target transaction complexity as the transaction to be learned by the branch robot.

[0204] Specifically, after determining the target transaction complexity, the present invention can determine the transaction to be evaluated corresponding to the target transaction complexity as the transaction to be learned by the branch robot.

[0205] Optionally, in other data processing devices proposed in this embodiment, the first sorting unit 103 is configured to execute: sorting the transaction complexity of each transaction to be evaluated in order from low to high; the second determination unit 104 is configured to execute: determining the transaction complexity corresponding to the sorting number that is not greater than a preset sequence number threshold as the target transaction complexity.

[0206] Among them, the present invention can sort the complexity of each transaction in order from low to high, and determine a certain number of sorting numbers in the front row as the target sequence numbers.

[0207] The preset sequence number threshold may be formulated by technical personnel according to actual working conditions, and the present invention does not limit this.

[0208] The data processing device proposed in this embodiment can establish an overall knowledge graph for each transaction to be evaluated in the transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated. According to the usage rate of the transaction elements, the difficulty of information extraction, and the number of training corpora, the transaction complexity of each transaction to be evaluated is determined respectively, and the transaction complexity of each transaction to be evaluated is sorted, and the transaction complexity corresponding to the sorting sequence number being the target sequence number is determined as the target transaction complexity, and the transaction to be evaluated corresponding to the target transaction complexity is determined as the transaction to be learned by the branch robot. The present invention can effectively determine the transaction to be learned by the branch robot.

[0209] based on Figure 5 The structural diagram shown is as follows: Figure 6 As shown, this embodiment proposes a second data processing device. In the device, the first determination unit 102 may include: a fourth determination unit 201, a fifth determination unit 202, a sixth determination unit 203 and a first triggering unit 204, wherein:

[0210] The fourth determining unit 201 is configured to execute: determining a transaction to be evaluated in the transaction set to be evaluated as a current target transaction to be evaluated;

[0211] It should be noted that the present invention can obtain the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated by successively determining the transactions to be evaluated in the transaction set to be evaluated as the current target transactions to be evaluated and determining the transaction complexity of the current target transactions to be evaluated.

[0212] The fifth determining unit 202 is configured to execute: determining the complexity of each transaction element in the current target transaction to be evaluated according to the usage rate of the transaction element, the difficulty of information extraction and the amount of training corpus;

[0213] Specifically, the present invention can determine the complexity of a transaction element according to the usage rate of the transaction element, the difficulty of information extraction and the amount of training corpus.

[0214] The sixth determining unit 203 is configured to execute: determining the average of the complexity of each transaction element in the current target transaction to be evaluated as the transaction complexity of the current target transaction to be evaluated;

[0215] Specifically, after determining the complexity of each transaction element in the current target transaction to be evaluated, the present invention can calculate the average of the complexity of each transaction element, and determine the calculated average as the transaction complexity of the current target transaction to be evaluated.

[0216] The first triggering unit 204 is configured to execute: triggering the fourth determining unit 201 until the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated is determined.

[0217] It should be noted that the present invention can execute the above process cyclically to determine the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated.

[0218] The data processing device proposed in this embodiment can obtain the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated by successively determining the transactions to be evaluated in the transaction set to be evaluated as the current target transactions to be evaluated and determining the transaction complexity of the current target transactions to be evaluated, thereby ensuring the success rate of obtaining the transaction complexity of each transaction to be evaluated.

[0219] based on Figure 6 As shown in the structural diagram, this embodiment proposes a third data processing device. In the device, the fifth determination unit 202 may include: a seventh determination unit, a usage rate determination unit, a difficulty determination unit, a corpus quantity determination unit, a complexity determination unit and a second trigger unit, wherein:

[0220] A seventh determining unit is configured to execute: determining a transaction element in the current target transaction to be evaluated as the current target transaction element;

[0221] The present invention can obtain the complexity of each transaction element in the current target transaction to be evaluated by sequentially determining each transaction element in the current target transaction to be evaluated as the current target transaction element and calculating the complexity of the current target transaction element.

[0222] The usage rate determination unit is configured to perform: determining the usage rate of the current target transaction element;

[0223] Optionally, the usage rate determination unit may include: determining the sum of the transaction volume proportions of each first transaction to be evaluated as the usage rate of the current target transaction element, and the first transaction to be evaluated is a transaction to be evaluated that includes the current target transaction element in the transaction set to be evaluated.

[0224] The transaction volume ratio of a certain transaction may be a ratio of the transaction volume of the transaction to the transaction volumes of all transactions to be evaluated in the transaction set to be evaluated.

[0225] Specifically, the present invention may determine the sum of the transaction volume proportions of each transaction to be evaluated that includes the current target transaction element as the utilization rate of the current target transaction element.

[0226] The difficulty determination unit is configured to perform: determining the difficulty of information extraction of the current target transaction element;

[0227] Optionally, the difficulty determination unit may include: a cosine distance calculation unit, a second sorting unit, an eighth determination unit, an average value calculation unit and a ninth determination unit, wherein:

[0228] The cosine distance calculation unit is configured to perform: calculating the cosine distances between the current target transaction element and other transaction elements in the overall knowledge graph;

[0229] The second sorting unit is configured to execute: sorting the calculated cosine distances in descending order;

[0230] An eighth determining unit is configured to execute: determining all cosine distances whose sorting numbers are not greater than a preset threshold as target cosine distances;

[0231] The average value calculation unit is configured to perform: calculating the average value of the cosine distances of each target;

[0232] The ninth determining unit is configured to execute: determining the reciprocal value of the calculated average value as the information extraction difficulty of the current target transaction element.

[0233] It is understandable that when the degree of distinction between a transaction element and other transaction elements is greater, the difficulty of extracting its information is smaller; when the degree of distinction between a transaction element and other transaction elements is smaller, the difficulty of extracting its information is greater. Among them, the present invention can determine the degree of distinction between a transaction element and other transaction elements based on the cosine distance between a transaction element and other transaction elements.

[0234] Among them, in the process of calculating the cosine distance between transaction elements, the present invention can first use the word vector model to perform text vectorization on each transaction element in the overall knowledge graph to obtain each transaction element represented by text vectorization. Then, for the two transaction elements whose cosine distance is to be calculated, the present invention calculates the cosine distance of the text vectors corresponding to the two transaction elements according to the cosine distance calculation formula between vectors to obtain the cosine distance between the two transaction elements.

[0235] Specifically, after calculating the cosine distance between the current target transaction element and other transaction elements in the overall knowledge graph, the cosine distances can be sorted in descending order, and a certain number of cosine distances in the front row can be determined as the target cosine distances. For example, the first 10% of the cosine distances can be determined as the target cosine distances. Afterwards, the present invention can determine the average value of the target cosine distance as the discrimination between the current target transaction element and other transaction elements, and determine the reciprocal of the discrimination as the information extraction difficulty of the current target transaction element.

[0236] The corpus quantity determination unit is configured to perform: determining the training corpus quantity of the current target transaction element;

[0237] Specifically, the present invention may determine the amount of training corpus for the current target transaction element used by the information extraction model in the training phase as the amount of training corpus for the current target transaction element.

[0238] The complexity determination unit is configured to perform: determining the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction and the amount of training corpus;

[0239] Specifically, the present invention may use a certain calculation formula to calculate the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction and the amount of training corpus.

[0240] The second triggering unit is configured to execute: triggering the seventh determining unit until the complexity of each transaction element in the current target transaction to be evaluated is determined.

[0241] Specifically, the present invention can execute the above process cyclically until the complexity of each transaction element in the current target transaction to be evaluated is determined.

[0242] The data processing device proposed in this embodiment can obtain the complexity of each transaction element in the current target transaction to be evaluated by determining each transaction element in the current target transaction to be evaluated as the current target transaction element in turn and calculating the complexity of the current target transaction element, thereby ensuring the success rate of obtaining the complexity of each transaction element in the current target transaction to be evaluated, thereby further ensuring the success rate of obtaining the transaction complexity of each transaction to be evaluated.

[0243] Based on the third data processing device described above, this embodiment proposes a fourth data processing device. In the fourth data processing device, the complexity determination unit may include: a first processing unit, a first obtaining unit, a second processing unit, a second obtaining unit, a third processing unit, a third obtaining unit, an input unit, and a fourth obtaining unit, wherein:

[0244] The first processing unit is configured to perform: normalizing the usage rate of the current target transaction element;

[0245] The first obtaining unit is configured to execute: obtaining a post-processing usage rate;

[0246] The second processing unit is configured to perform: normalizing the difficulty of information extraction of the current target transaction element;

[0247] The second obtaining unit is configured to execute: obtaining the difficulty of extracting the processed information;

[0248] The third processing unit is configured to perform: normalizing the number of training corpora of the current target transaction element;

[0249] The third obtaining unit is configured to execute: obtaining the amount of processed training corpus;

[0250] The input unit is configured to execute: inputting the post-processing usage rate, the post-processing information extraction difficulty and the post-processing training corpus quantity into the complexity calculation formula:

[0251] S=r+n-df

[0252] In which, S is the complexity of the current target transaction element, r is the utilization rate of the current target transaction element after processing, n is the difficulty of information extraction after processing of the current target transaction element, and df is the number of training corpora after processing of the current target transaction element;

[0253] The fourth obtaining unit is configured to execute: obtaining the complexity of the current target transaction element output by the complexity calculation formula.

[0254] It can be understood that in the process of calculating complexity based on usage rate, information extraction difficulty and the number of training corpora, in order to effectively improve the accuracy of quantitative measurement of complexity, the present invention can first apply normalization processing technology to normalize the usage rate, information extraction difficulty and the number of training corpora respectively, and then use the normalized usage rate, information extraction difficulty and the number of training corpora to calculate the complexity.

[0255] Among them, after obtaining the usage rate, information extraction difficulty and number of training corpora of each transaction element in the overall knowledge graph, the present invention can perform normalization processing on the usage rate, information extraction difficulty and number of training corpora of each transaction element respectively, and map the values ​​of usage rate, information extraction difficulty and number of training corpora to the range of 0 to 1.

[0256] Specifically, the present invention can calculate the complexity of the current target transaction element according to the complexity calculation formula after obtaining the post-processing utilization rate, post-processing information extraction difficulty and post-processing training corpus quantity of the current target transaction element.

[0257] The data processing device proposed in this embodiment can first apply normalization processing technology to normalize the usage rate, information extraction difficulty and the number of training corpora respectively, and then use the normalized usage rate, information extraction difficulty and the number of training corpora to calculate the complexity, thereby effectively improving the accuracy of quantitative calculation of the complexity of transaction elements.

[0258] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0259] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A data processing method, characterized in that: include: For each transaction to be evaluated in the transaction set to be evaluated, an overall knowledge graph is established, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated; Determine one of the transactions to be evaluated in the transaction set to be evaluated as a current target transaction to be evaluated; Determine a transaction element in the current target transaction to be evaluated as the current target transaction element; Determine the current utilization rate of target transaction elements; Calculating the cosine distance between the current target transaction element and other transaction elements in the overall knowledge graph; The calculated cosine distances are sorted in descending order, and the cosine distances whose sorting numbers are not greater than a preset threshold are determined as target cosine distances; Calculating the average of the target cosine distances; Determine the calculated reciprocal value of the average value as the information extraction difficulty of the current target transaction element; Determine the amount of training corpus for the current target transaction element; Determine the complexity of the current target transaction elements based on the usage rate of the current target transaction elements, the difficulty of information extraction, and the amount of training corpus; Returning to the step of determining a transaction element in the current target transaction to be evaluated as the current target transaction element, until the complexity of each transaction element in the current target transaction to be evaluated is determined; The average of the complexity of each transaction element in the current target transaction to be evaluated is determined as the transaction complexity of the current target transaction to be evaluated; Returning to the step of determining one of the transactions to be evaluated in the transaction set to be evaluated as the current target transaction to be evaluated, until the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated is determined; Sorting the transaction complexity of each of the transactions to be evaluated, and determining the transaction complexity corresponding to the target sequence number as the target transaction complexity; Determine the transaction to be evaluated corresponding to the target transaction complexity as the transaction to be learned by the branch robot; The step of determining the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction, and the amount of training corpus includes: Normalize the usage rate of the current target transaction element to obtain the processed usage rate; Normalizing the difficulty of information extraction of the current target transaction element to obtain the difficulty of information extraction after processing; Normalize the number of training corpora of the current target transaction element to obtain the number of training corpora after processing; The post-processing utilization rate, the post-processing information extraction difficulty and the post-processing training corpus quantity are input into the complexity calculation formula: S=r+n-df In which, S is the complexity of the current target transaction element, r is the utilization rate of the current target transaction element after processing, n is the difficulty of information extraction after processing of the current target transaction element, and df is the number of training corpora after processing of the current target transaction element; The complexity of the current target transaction element output by the complexity calculation formula is obtained.

2. The method according to claim 1, characterized in that The overall knowledge graph is established for each transaction to be evaluated in the transaction set to be evaluated, including: Establishing knowledge graphs for the transactions to be evaluated respectively; Based on the upstream and downstream relationships between the transactions to be evaluated, the overall knowledge graph is established using the knowledge graphs of the transactions to be evaluated.

3. The method according to claim 1, characterized in that The step of ranking the transaction complexity of each transaction to be evaluated includes: Sorting the transaction complexity of each transaction to be evaluated from low to high; The step of determining the transaction complexity corresponding to the target sequence number as the target transaction complexity includes: The transaction complexity corresponding to the sorting sequence number that is not greater than the preset sequence number threshold is determined as the target transaction complexity.

4. The method according to claim 1, characterized in that: The determining of the current usage rate of the target transaction element includes: The sum of the transaction volume proportions of the first transactions to be evaluated is determined as the usage rate of the current target transaction element, wherein the first transactions to be evaluated are the transactions to be evaluated that include the current target transaction element in the transaction set to be evaluated.

5. A data processing device, characterized in that: include: A first establishing unit, a first determining unit, a first sorting unit, a second determining unit and a third determining unit, wherein: The first establishing unit is configured to execute: establishing an overall knowledge graph for each transaction to be evaluated in the transaction set to be evaluated, wherein the transaction set to be evaluated includes at least one transaction to be evaluated, and the overall knowledge graph includes transaction elements of each transaction to be evaluated; The first determination unit is configured to execute: determining the transaction complexity of each of the transactions to be evaluated according to the usage rate of transaction elements, the difficulty of information extraction and the amount of training corpus; The first sorting unit is configured to execute: sorting the transaction complexity of each of the transactions to be evaluated; The second determining unit is configured to execute: determining the transaction complexity corresponding to the target sequence number as the target transaction complexity; The third determining unit is configured to execute: determining the transaction to be evaluated corresponding to the target transaction complexity as a transaction to be learned by the branch robot; The first determining unit includes: a fourth determining unit, a fifth determining unit, a sixth determining unit and a first triggering unit, wherein: The fourth determining unit is configured to execute: determining one of the transactions to be evaluated in the set of transactions to be evaluated as a current target transaction to be evaluated; The fifth determining unit is configured to execute: determining the complexity of each transaction element in the current target transaction to be evaluated according to the usage rate of the transaction element, the difficulty of information extraction and the amount of training corpus; The sixth determining unit is configured to execute: determining the average of the complexity of each transaction element in the current target transaction to be evaluated as the transaction complexity of the current target transaction to be evaluated; The first triggering unit is configured to execute: triggering the fourth determining unit until the transaction complexity of each transaction to be evaluated in the transaction set to be evaluated is determined; The fifth determination unit includes: a seventh determination unit, a usage rate determination unit, a difficulty determination unit, a corpus quantity determination unit, a complexity determination unit, and a second triggering unit, wherein: The seventh determining unit is configured to execute: determining a transaction element in the current target transaction to be evaluated as the current target transaction element; The usage rate determination unit is configured to perform: determining the usage rate of the current target transaction element; The difficulty determination unit is configured to perform: determining the difficulty of information extraction of the current target transaction element; The corpus quantity determination unit is configured to perform: determining the training corpus quantity of the current target transaction element; The complexity determination unit is configured to perform: determining the complexity of the current target transaction element according to the usage rate of the current target transaction element, the difficulty of information extraction and the amount of training corpus; The second triggering unit is configured to execute: triggering the seventh determining unit until the complexity of each transaction element in the current target transaction to be evaluated is determined; The difficulty determination unit includes: a cosine distance calculation unit, a second sorting unit, an eighth determination unit, an average value calculation unit and a ninth determination unit, wherein: The cosine distance calculation unit is configured to perform: calculating the cosine distances between the current target transaction element and other transaction elements in the overall knowledge graph; The second sorting unit is configured to execute: sorting the calculated cosine distances in descending order; The eighth determination unit is configured to execute: determining all cosine distances whose sorting numbers are not greater than a preset threshold as target cosine distances; The average value calculation unit is configured to perform: calculating the average value of each of the target cosine distances; The ninth determining unit is configured to execute: determining the calculated reciprocal value of the average value as the information extraction difficulty of the current target transaction element; The complexity determination unit includes: a first processing unit, a first obtaining unit, a second processing unit, a second obtaining unit, a third processing unit, a third obtaining unit, an input unit and a fourth obtaining unit, wherein: The first processing unit is configured to perform: normalizing the usage rate of the current target transaction element; The first obtaining unit is configured to execute: obtaining a post-processing usage rate; The second processing unit is configured to perform: normalizing the difficulty of information extraction of the current target transaction element; The second obtaining unit is configured to perform: obtaining a difficulty of extracting processed information; The third processing unit is configured to perform: normalizing the number of training corpora of the current target transaction element; The third obtaining unit is configured to execute: obtaining the amount of processed training corpus; The input unit is configured to execute: inputting the post-processing utilization rate, the post-processing information extraction difficulty and the post-processing training corpus quantity into the complexity calculation formula: S=r+n-df In which, S is the complexity of the current target transaction element, r is the utilization rate of the current target transaction element after processing, n is the difficulty of information extraction after processing of the current target transaction element, and df is the number of training corpora after processing of the current target transaction element; The fourth obtaining unit is configured to execute: obtaining the complexity of the current target transaction element output by the complexity calculation formula.

6. The device according to claim 5, characterized in that The first establishing unit includes: a second establishing unit and a third establishing unit, wherein: The second establishing unit is configured to execute: establishing a knowledge graph of each of the transactions to be evaluated respectively; The third establishing unit is configured to execute: based on the upstream and downstream relationships between the transactions to be evaluated, using the knowledge graphs of the transactions to be evaluated to establish the overall knowledge graph.

Images