Complaint processing method and device, readable storage medium and electronic equipment

Abstract

The invention discloses an appeal processing method and device, a readable storage medium and electronic equipment, and the method comprises the steps: obtaining an input text returned by a user for an output text in response to an appeal request, matching the text features of the input text with the element features of each risk element node contained in a risk control map, and determining a target element node. And determining at least one historical risk event node corresponding to the target element node, and determining the risk degree according to the weight of the edge between the at least one historical risk event node and the target element node. And according to the risk degree, determining whether the target service can be determined as an abnormal service. And if yes, not delimiting the target service. And if not, delimiting the target service. According to the method, whether the target service is the abnormal service or not is further judged according to the connection relation between the nodes in the risk control graph and the weight of the edge, automatic processing of the user appeal is achieved, and compared with a manual delimitation mode, the processing efficiency of the user appeal is greatly improved.

Description

Technical Field

[0001] The present specification relates to the field of computer technology, and in particular to a complaint handling method, device, readable storage medium, and electronic device. Background Art

[0002] Nowadays, the application of artificial intelligence is becoming more and more widespread. In Internet business, risk assessment can be carried out on the business through risk control models based on artificial intelligence training, abnormal business can be identified and corresponding restriction measures can be taken on the business to protect the user's private data.

[0003] However, due to the diversity and uncertainty of user behavior, the accuracy of business risk assessment is affected. It is possible that normal business is mistakenly identified as abnormal business, and restrictions are imposed on normal business. At this time, if the user wants to lift the restrictions, he or she generally needs to file a complaint with the manual customer service, who will understand the specific business details, further determine whether the business is misidentified, and lift the restrictions on the misidentified business.

[0004] It can be seen that the existing method of removing restriction measures has low processing efficiency and long waiting time for users. Therefore, this specification provides a complaint handling method. Summary of the invention

[0005] This specification provides a complaint handling method, device, readable storage medium and electronic device to partially solve the above-mentioned problems existing in the prior art.

[0006] This manual adopts the following technical solutions:

[0007] This manual provides a method for handling complaints, including:

[0008] In response to a user's complaint request for a target service, a preset output text is sent to the user, and an input text returned by the user to the output text is obtained;

[0009] Determine the text features of the input text, match the text features with the feature features of each risk factor node included in the risk control map, and determine the target factor node; the risk control map is composed of historical risk event nodes and risk factor nodes, and edges connecting the historical risk event nodes and the risk factor nodes, and the weight of the edge represents the influence of the risk factor on the historical risk event;

[0010] Determine at least one historical risk event node corresponding to the target element node, and determine the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node;

[0011] Determining whether the target business can be determined as an abnormal business according to the risk level;

[0012] If yes, determine that the target service is an abnormal service, and do not unblock the target service;

[0013] If not, it is determined that the target business is a normal business, and the target business is unrestricted.

[0014] Optionally, the risk control map is determined by the following method, wherein:

[0015] Acquire description texts of multiple abnormal historical services, and extract events from each description text to determine event keywords and event elements of the multiple abnormal historical services;

[0016] Cluster each abnormal historical business according to the determined event keywords;

[0017] For each cluster, according to the event elements of each abnormal historical service contained in the cluster, the event elements of the cluster are deduplicated, and according to the event keywords of each abnormal historical service contained in the cluster, the representation keywords of the cluster are determined;

[0018] According to the representative keywords of each cluster, determine the historical risk event nodes corresponding to each cluster, and according to the event elements after deduplication of each cluster, determine the risk factor nodes corresponding to each historical risk event node;

[0019] For each risk factor node corresponding to each historical risk event node, determine the ratio of the number of occurrences of the event element corresponding to the risk factor node in each abnormal historical business included in the cluster corresponding to the historical risk event node to the total number of occurrences in each abnormal historical business, and determine the weight of the edge between the risk factor node and the historical risk event node according to the ratio;

[0020] For each historical risk event node, the weights of the edges connecting the historical risk event node are normalized to obtain the risk control graph.

[0021] Optionally, the method further comprises:

[0022] When the target business cannot be determined as an abnormal business according to the risk level, continuing to determine an output text to be sent to the user according to the risk factor nodes connected to the at least one historical risk event node;

[0023] Acquire the input text returned by the user to the re-determined output text, and update the risk level according to the returned input text;

[0024] It is further determined whether the target business can be determined as an abnormal business based on the updated risk level.

[0025] Optionally, continuing to determine the output text to be sent to the user according to the risk factor node connected to the at least one historical risk event node specifically includes:

[0026] Determining whether there are any risk factor nodes that have not been visited among the risk factor nodes connected to the at least one historical risk event node;

[0027] If yes, determine a pending element node among the risk element nodes that have not been visited and are connected to the at least one historical risk event node; and continue to determine the output text to be sent to the user according to the pending element node;

[0028] If not, it is determined that all risk factor nodes connected to the at least one historical risk event node have been visited.

[0029] Optionally, determining the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node specifically includes:

[0030] Determine a target event node from at least one historical risk event that has not been visited and corresponds to the target element node according to the weight of the edge between the at least one historical risk event and the target element node;

[0031] The risk level of the target business is updated according to the weight of the edge between the target element node and the target event node.

[0032] Optionally, determining the undetermined element node among the risk element nodes that have not been visited and are connected to the at least one historical risk event node specifically includes:

[0033] According to the weights of the edges corresponding to the at least one historical risk event node, a pending element node is determined from among the unvisited risk element nodes connected to the at least one historical risk event node.

[0034] Optionally, updating the risk level according to the input text returned again specifically includes:

[0035] Determine the text features of the input text returned again, match the text features of the input text returned again with the element features of the risk element corresponding to the undetermined element node, and determine a first similarity;

[0036] Determining whether the first similarity is greater than a preset first threshold;

[0037] If so, determine that the target business matches the undetermined element node, and cumulatively update the risk level through the weight of the edge between the at least one historical event node and the undetermined element node;

[0038] If not, it is determined that the target business does not match the undetermined element node, and the risk level is not updated.

[0039] This specification provides a complaint handling device, including:

[0040] A response module, in response to a user's complaint request for a target service, sends a preset output text to the user, and obtains an input text returned by the user to the output text;

[0041] a target element node determination module, which determines the text features of the input text, matches the text features with the element features of each risk element node included in the risk control map, and determines the target element node; the risk control map is composed of historical risk event nodes and risk element nodes, and edges connecting historical risk event nodes and risk element nodes, and the weight of the edge represents the influence of the risk element on the historical risk event;

[0042] a risk level determination module, which determines at least one historical risk event node corresponding to the target element node, and determines the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node;

[0043] A risk identification module, which determines whether the target business can be determined as an abnormal business according to the risk level;

[0044] A normal business determination module determines that the target business is an abnormal business and does not remove the restriction on the target business.

[0045] The abnormal business determination module determines that the target business is a normal business, if not, and removes the restriction on the target business.

[0046] The computer-readable storage medium provided in this specification stores a computer program, and when the computer program is executed by a processor, any of the above-mentioned complaint handling methods is implemented.

[0047] The electronic device provided in this specification includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned complaint handling methods when executing the program.

[0048] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:

[0049] In the complaint handling method provided in this specification, in response to the user's complaint request for the target business, the input text returned by the user to the preset output text is obtained, and the text features of the input text are matched with the element features of each risk element node contained in the risk control map to determine the target element node. The risk control map is composed of historical risk event nodes and risk element nodes, and edges connecting historical risk event nodes and risk element nodes. The weight of the edge represents the influence of the risk factor on the historical risk event. At least one historical risk event node corresponding to the target element node is determined, and the risk level of the target business is determined according to the weight of the edge between at least one historical risk event node and the target element node. According to the risk level, it is judged whether the target business can be determined as an abnormal business. If so, the target business is determined to be an abnormal business, and the target business is not unrestricted. If not, the target business is determined to be a normal business, and the target business is unrestricted.

[0050] It can be seen from the above method that in this method, the target element node is determined through the input text returned by the user, and the target business is connected to the risk control map through the target element node, so as to further determine whether the target business is an abnormal business by utilizing the connection relationship between the nodes in the risk control map and the weight of the edge, thereby realizing automatic processing of user complaints. Compared with the manual release method, the efficiency of handling user complaints is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The illustrative embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation on this specification. In the drawings:

[0052] Figure 1 A flowchart of the complaint handling method provided for this manual;

[0053] Figure 2 A partial schematic diagram of a risk control map provided in an embodiment of this specification;

[0054] Figure 3 This is a flowchart of the specific implementation of step S110 of the complaint handling method provided in this specification;

[0055] Figure 4 A method provided in the embodiments of this specification Figure 3 The step S200 is shown as a specific implementation flow chart;

[0056] Figure 5 A method provided in the embodiments of this specification Figure 3 The step S202 is shown as a specific implementation flow chart;

[0057] Figure 6A partial schematic diagram of another wind control map provided in an embodiment of this specification;

[0058] Figure 7 A schematic diagram of the structure of the complaint handling device provided in this manual;

[0059] Figure 8 The corresponding Figure 1 Schematic diagram of electronic equipment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0061] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0062] Figure 1 The following is a flowchart of the complaint handling method provided in this manual, which specifically includes the following steps:

[0063] S100: In response to a user's complaint request for a target service, a preset output text is sent to the user, and an input text returned by the user to the output text is obtained.

[0064] All steps in the complaint handling method provided in this specification can be implemented by any electronic device with computing functions, such as a terminal, a server, etc. For ease of description, the complaint handling method provided in this specification is described below with the server as the execution subject.

[0065] In the complaint handling method provided in this specification, when a user initiates a complaint request, the server sends a preset output text to the user and obtains the input text returned by the user to the output text.

[0066] The preset output text is used to guide the user to input business information, through which the target business can be further judged whether it is an abnormal business. The preset output text is usually set based on experience to guide the user to input questions related to the risk level of the target business, such as questions about the purpose of the transaction, the identity of the transaction counterparty, etc.

[0067] S102: Determine the text features of the input text, match the text features with the feature features of each risk factor node included in the risk control map, and determine the target factor node; the risk control map is composed of historical risk event nodes and risk factor nodes, and edges connecting historical risk event nodes and risk factor nodes, and the weight of the edge represents the influence of the risk factor on the historical risk event.

[0068] In the complaint handling method provided in this specification, the output text sent to the user during the conversation is determined by the connection relationship of the nodes in the risk control map. The risk control map consists of historical risk event nodes and risk factor nodes, and edges connecting historical risk event nodes and risk factor nodes. The risk control map is a weighted graph, and the weight of the edge represents the influence of the risk factor on the historical risk event.

[0069] For the interconnected historical risk event nodes and risk factor nodes, the degree of influence between the historical risk event corresponding to the historical risk event node and the risk factor corresponding to the risk factor node indicates the degree of influence of the existence of the risk factor corresponding to the risk factor node in the historical risk event on the result of determining that the historical risk event is an abnormal business with risks.

[0070] The server needs to connect the input text returned by the user with the nodes in the risk control map. Therefore, the server pre-encodes the risk factors corresponding to the risk factor nodes in the risk control map through the trained word embedding model to determine the factor characteristics of each risk factor node.

[0071] When the input text returned by the user is obtained, the server encodes the input text through the word embedding model to determine the text features of the input text. Then, the server matches the text features of the input text with the feature features corresponding to each risk factor node to determine the target feature node. The word embedding model here can be implemented by any embedding representation model for text, such as the bidirectional encoder representation model based on transformers (Bidirectional Encoder Representations from Transformers, BERT), bag of words model (Bag of Words, BOW), global vectors for word representation model based on global word frequency statistics (GloVe), etc.

[0072] During matching, the server can respectively determine the second similarity between the element features of each risk element node and the text features of the input text, and screen each risk element node according to the second similarity corresponding to the element features of each risk element node through a preset second threshold to determine the target element node.

[0073] The specific meaning of the second threshold is not limited in this specification, and it can be set according to different screening methods. For example, the second threshold can be set as a screening value of similarity, and the server will use the risk factor node corresponding to the second similarity greater than the second threshold as the target factor node. The second threshold can also be set as a screening value of quantity, and the server will sort the risk factor nodes in descending order of the second similarities, and will determine the risk factor nodes with the second threshold number ranked first as the target factor node.

[0074] It can be seen that the number of target element nodes is related to the specific numerical setting of the second threshold, and the number of target element nodes obtained after matching may be one or more. Below, the complaint handling method provided in this specification is described by taking the number of target element nodes as one as an example.

[0075] S104: Determine at least one historical risk event node corresponding to the target element node, and determine the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node.

[0076] In the risk control graph, a risk factor node may be connected to multiple historical risk event nodes. After determining the target factor node, the server determines at least one historical risk event node corresponding to the target factor node, and determines the risk level of the target business based on the weight of the edge between the at least one historical risk event node and the target factor node.

[0077] When the target element node is connected to only one historical risk event node, the server determines the risk event that the target business may correspond to based on this historical risk event node, and subsequently judges the risk level of the target business based on this historical risk event node. When the target element node is connected to multiple historical risk event nodes, the server determines the risk event that the target business may correspond to based on multiple risk event nodes, and subsequently judges the risk level of the target business based on multiple historical risk event nodes.

[0078] In the complaint handling method provided in this specification, only when the server determines that the target business does not have a corresponding relationship with all the historical risk event nodes connected to the target element node, will it determine that the target business is a misidentified business, that is, a normal business, and unblock the target business. As long as the server can determine that the target business has a corresponding relationship with any historical risk event node connected to the target element node, the target business will not be unblocked.

[0079] The risk levels of different historical risk event nodes need to be calculated separately so that the server can determine which historical risk event node the target business corresponds to. There are two ways to determine the risk levels of different historical risk event nodes, one is to determine them one by one, and the other is to determine them simultaneously.

[0080] The idea of ​​determining in sequence is that each time the server determines a target event node in at least one historical risk event node corresponding to the target element node, and each time only determines the risk level corresponding to the target event node for this target event node.

[0081] In one or more embodiments of the present specification, the server determines the risk level of each historical risk event in turn. The server determines a historical risk event node from the at least one historical risk event node that has not been visited corresponding to the target element node as the target event node, and the target event node is the historical risk event node currently visited by the server. The risk level of the target business is determined based on the weight of the edge between the target element node and the target event node. The risk level is used to determine whether the target business has the risk corresponding to the target event node, and is the risk level corresponding to the currently visited historical risk event node.

[0082] The same historical risk event node only needs to be visited once to determine the risk level once. Therefore, when the target element node corresponds to multiple historical risk event nodes, it is necessary to redetermine the target event node among the historical risk event nodes that have not been visited.

[0083] In this embodiment, there is no restriction on the method of determining the target event node. When the server determines the target event node, it can be randomly selected. That is, the server randomly selects a historical risk event node from the at least one historical risk event node that has not been visited corresponding to the target element node as the target event node.

[0084] In order to speed up the complaint handling process for the target business, the server may also select the target event node according to the weight. That is, the server determines the target event node from at least one historical risk event node that has not been visited and corresponds to the target element node, according to the weight of the edge between the target element node and the at least one historical risk event.

[0085] Specifically, the server may use the historical risk event node corresponding to the edge with the largest weight as the target event node among at least one historical risk event node that has not been visited corresponding to the target element node. Because the risk level corresponding to the current target event node is determined based on the weight of the edge between the target element node and the target event node, if the risk level determined based on the edge with the largest weight can determine that the target business is an abnormal business, it means that the server has determined that there is a corresponding relationship between the target business and the target event node, so there is no need to visit other historical risk event nodes to determine the appeal result, thereby speeding up the execution rate of ending the appeal process.

[0086] The following will be combined Figure 2 The method of this embodiment is described. Figure 2 This is a partial schematic diagram of a risk control map provided in an embodiment of this specification. Rectangles represent historical risk event nodes, circles represent risk factor nodes, and circles with shadows represent target factor nodes. Figure 2 In the relevant description, the letters in the rectangles refer to the historical risk event nodes, the numbers in the circles refer to the risk factor nodes, and the letters on the edges represent the weights of the edges. Figure 2 As shown, the target element node "3" corresponds to three historical risk event nodes, namely "a", "b" and "c". If the currently visited historical risk event node is "a", then "a" is the target event node. The server determines that the risk level corresponding to "a" is w3 based on the weight of the edge between "a" and "3".

[0087] S106: According to the risk level, determine whether the target business can be determined as an abnormal business, if so, execute S108, if not, execute S110.

[0088] S108: Determine that the target service is an abnormal service, and do not remove restrictions on the target service.

[0089] S110: Determine that the target service is a normal service, and unblock the target service.

[0090] The server will determine whether the target business has a corresponding relationship with any historical risk event node based on the input text returned by the user. Each time the risk level is determined for a historical risk event node, it is determined whether the target business has a corresponding relationship with the currently determined target event node. The risk level here refers to the risk level corresponding to the target event node, that is, the risk level corresponding to the currently visited historical risk event node.

[0091] The server determines whether the risk level corresponding to the target event node is greater than a preset risk threshold. When the risk level corresponding to the target event node is not greater than the preset risk threshold, the server cannot determine that there is a corresponding relationship between the target business and the target event node.

[0092] When the risk level corresponding to the target event node is greater than the preset risk threshold, the server can determine that there is a corresponding relationship between the target business and the target event node, that is, the target business is an abnormal business. In this case, the complaint result can be returned to the user without lifting the limit on the target business.

[0093] In order to further warn the user, the server can determine the historical risk event node corresponding to the risk level greater than the risk threshold, and generate the case text corresponding to the historical risk event node according to the risk factor nodes connected to the historical risk event node corresponding to the risk level. The case text is similar to the risk status of the target business, which can make the user more clearly aware of the risk status of the target business.

[0094] It can be seen from the above method that in this method, the target element node is determined through the input text returned by the user, and the target business is connected to the risk control map through the target element node, so as to further determine whether the target business is an abnormal business by utilizing the connection relationship between the nodes in the risk control map and the weight of the edge, thereby realizing automatic processing of user complaints. Compared with the manual release method, the efficiency of handling user complaints is greatly improved.

[0095] In the above step S102, the server needs to determine the target element node in the risk control map based on the input text returned by the user for the first time before continuing to execute steps S104 to S110 to determine the risk level of the target business and determine whether to lift the limit on the target business based on the risk level.

[0096] However, in some cases, when the second threshold is a screening value for similarity, it is possible that the second similarity between the element features of all risk element nodes in the risk control map and the text features of the input text returned by the user for the first time is not greater than the second threshold. Then the server cannot determine the target element node, and cannot establish a connection between the target business and the risk control map, resulting in the inability to execute subsequent steps.

[0097] Therefore, when the server cannot determine the target element node in S102, the server directly determines that the target business is at risk, does not remove the restrictions on the target business, and returns the appeal result of "determined to have a risk" to the user.

[0098] Because in the context of the complaint in this specification, the target business has been identified as a potentially abnormal business with risks by the risk control model trained by artificial intelligence. When the user complains, no information related to the risk level of the target business is returned, resulting in the server obtaining a result of whether the target business has risks based on the risk level. However, in order to ensure the security of the business environment and user data, no potential abnormal business can be missed, and the server does not remove the restrictions on the target business.

[0099] However, in order to make this method more flexible in dealing with situations where the target element node cannot be determined, the rate of unblocking the misidentified business can be further improved. If the appeal result of "risk is determined to exist" is returned because the target element node cannot be determined, the server continues to receive the input text returned by the user during the additional waiting time. Users who do not agree with the appeal result can continue to input text, and the server continues to determine the text features of the text input by the user again during the waiting time, and matches the text features with the element features of each risk element node in the risk control map again according to the text features to determine the target element node. As long as the input text returned by the user during the waiting time can be linked to the risk element node in the risk control map, the above steps S104 to S110 can be executed to accurately determine whether to unblock the target business.

[0100] If within the waiting time, the server still cannot determine the target feature node based on the input text returned by the user, the server determines that the target business is at risk and does not remove the restriction on the target business.

[0101] In order to achieve the same purpose, in another embodiment, there can be multiple preset input texts. When in S102, the server cannot determine the target element node according to the input text returned by the user for the first preset output text, the server selects an output text that has not been sent to the user in the preset output text, and obtains the input text returned by the user again. Then, the server matches the text features of the input text returned by the user again with the feature features of each risk element node in the risk control map again, and determines the second similarity of each feature with the text features of the input text returned by the user again, and determines whether there is at least one second similarity greater than the second threshold. If so, the server uses the risk element node corresponding to the second similarity greater than the second threshold as the target element node, and continues to perform the above steps S104 to S110. If not, continue to select an output text that has not been sent to the user in the preset output text, until all preset output texts are sent, and the target element node cannot be determined, the server determines that the target business is at risk and does not release the limit on the target business.

[0102] The risk control map mentioned in the above step S102 can be constructed as follows.

[0103] The server needs to obtain the description texts of multiple abnormal historical services, and extract events from each description text to determine the event keywords and event elements of each abnormal historical service. There are many mature methods for event extraction, such as feature-based methods, neural network-based methods, etc. This is not the focus of this manual and will not be described in detail.

[0104] In one or more embodiments of the present specification, the server clusters each abnormal historical business according to the determined event keywords, and obtains each clustering cluster of each abnormal historical business classified according to the semantics of the event keywords. The abnormal historical business contained in each clustering cluster can be considered as a similar business, and the same event elements are likely to exist in the different abnormal businesses contained in the clustering cluster. In this embodiment, the abnormal historical business contained in a clustering cluster will be used to construct a subgraph of the risk control map, and the event elements contained in the clustering cluster will be used to determine the event element nodes in the subgraph, and the same event element nodes only need to appear once in the subgraph. The server needs to deduplicate the event elements contained in each clustering cluster according to the event elements of each abnormal historical business contained in the clustering cluster, and determine the remaining event elements of the clustering cluster.

[0105] Then, the server determines the representative keyword of the cluster according to the event keywords of each abnormal historical business included in the cluster. There are many ways to determine the representative keyword, for example, the server selects the event keyword closest to the cluster center of the cluster, or randomly selects from the event keywords included in the cluster.

[0106] The server determines the historical risk event node of the cluster according to the representative keywords of the cluster, and determines the risk factor nodes corresponding to the historical risk event node of the cluster according to the event elements of the cluster after deduplication.

[0107] After determining the risk factor nodes corresponding to all historical risk event nodes, the server uses, for each risk factor node corresponding to each historical risk event node, the number of occurrences of the event element corresponding to the risk factor node in the event elements of each abnormal historical business contained in the cluster corresponding to the historical risk event node as a first value, and the total number of occurrences of the event element corresponding to the risk factor node in the event elements of each abnormal historical business as a second value, and determines the weight of the edge between the risk factor node and the historical risk event node based on the ratio of the first value to the second value.

[0108] Finally, for each historical risk event node, the risk control map is determined according to the weights of each edge connected to the historical risk event node.

[0109] Furthermore, in order to unify the weight scale of each edge connected to each historical risk event node so that the weight has the same measurement standard for the impact of different historical risk events, the server can normalize the weight of each edge connected to the historical risk event node. Based on the normalized edges, the risk control map is determined.

[0110] In this embodiment, the weight of the edge between the historical risk event node and the risk factor node is determined according to the ratio of the first value to the second value. The ratio can be used to indicate the degree of association of the occurrence of the historical risk event when the event element corresponding to the risk factor node occurs. The larger the ratio, the greater the possibility that the event element corresponding to the risk factor node occurs and the abnormal historical business corresponding to the historical risk event node occurs, that is, the risk factor node has a high degree of influence on the historical risk event node. The smaller the ratio, the smaller the possibility that the event element corresponding to the risk factor node occurs and the abnormal historical business corresponding to the historical risk event node occurs, that is, the risk factor node has a low degree of influence on the historical risk event node.

[0111] Of course, other methods can be used to determine the weight of the edge between the historical risk event node and the risk factor node. Since the event element corresponding to the abnormal historical business can itself be used as a risk identification indicator, if an event element appears frequently in various abnormal historical businesses, it can be said that the event element has a greater impact on identifying a business as a risky business.

[0112] Therefore, the server can also determine the weight of the edge corresponding to each risk factor node according to the number of occurrences of the event element corresponding to each risk factor node in the event element of each abnormal historical business. After determining the weights of all edges, a normalization operation is performed for each historical risk factor node.

[0113] In one or more embodiments of the present specification, after determining the event keywords and event elements of each abnormal historical business, the server determines, for each abnormal historical business, the degree of influence of the event elements of the abnormal historical business and the result that the abnormal historical business is determined as an abnormal business through a risk prediction model.

[0114] Specifically, the server inputs each event element of the abnormal historical business into the risk prediction model to determine the initial confidence level that the abnormal historical business is determined to be an abnormal business.

[0115] Secondly, for each event element of the abnormal historical business, the server deletes the event element from the event elements of the abnormal historical business, inputs the remaining event elements into the risk prediction model, and determines the control confidence of the abnormal historical business being determined as abnormal business. The difference between the initial confidence and the control confidence can indicate the degree of influence of the event element on the result that the abnormal historical business is determined as abnormal business.

[0116] Then, the server determines the historical risk event node of the abnormal historical business according to the event keyword of the abnormal historical business, and determines the risk element node corresponding to the historical risk event node of the abnormal historical business according to the event element. The difference between the initial confidence and the control confidence is used as the weight of the edge between the historical risk event node of the abnormal historical business and the risk element node.

[0117] After determining the weights of the edges between the historical risk event node of the abnormal historical business and all the risk element nodes of the abnormal historical business, a risk control map is obtained. Similarly, in order to unify the weight scale of each edge connected to each historical risk event node, the server here can also normalize the weights of each edge connected to the historical risk event node.

[0118] As mentioned in the above step S104, the server can determine the risk level according to the idea of ​​simultaneous determination. In this embodiment, the specific steps of the idea of ​​simultaneous determination will be introduced.

[0119] The idea of ​​simultaneous determination is that when determining the risk level, the server takes all risk factor nodes connected to at least one historical risk event node corresponding to the target factor node as a whole, and determines the risk level of the historical risk event node corresponding to each risk factor node in turn.

[0120] If the idea of ​​simultaneous determination is followed, then in the above step S104, after determining the target element node, the server determines the risk level of each historical risk event node according to at least one historical risk event node corresponding to the target element node.

[0121] Specifically, when the target element node is connected to only one historical risk event node, the risk level of the target business is determined according to the weight of the edge between the target element node and the historical risk event node.

[0122] When the target element node is connected to at least two historical risk event nodes, the server simultaneously determines the risk level corresponding to each historical event node based on the weight of the edge between each historical event node and the target element node.

[0123] Still Figure 2For example, in this embodiment, the server determines that the risk level corresponding to "a" is w3, the risk level corresponding to "b" is w4, and the risk level corresponding to "c" is w7 based on the weights of the edges between the target element node "3" and "a", "b" and "c".

[0124] In this embodiment, the risk level refers to the risk level of all historical risk event nodes corresponding to the target element node, which may be one or more. In the above step S106, when the server determines whether the target business can be determined as an abnormal business based on the risk level, it actually determines whether there is any risk level greater than the preset risk threshold among the risk levels of all historical risk event nodes corresponding to the target element node. As long as it is determined that there is any risk level greater than the preset risk threshold among the risk levels of all historical risk event nodes corresponding to the target element node, it means that the target business can be determined as an abnormal business.

[0125] In the method described in S100 to S110 above, the server determines the appeal result of the target business through only one round of dialogue. That is, the risk level of the target business is determined only based on an input text returned by the user. When determining the risk level, for a historical risk event node, only the weight of one edge between the historical risk event node and the target element node is considered. Under normal circumstances, a historical risk event node in the risk control map is connected to multiple risk element nodes, and the risk elements corresponding to the multiple risk element nodes connected to the historical risk event node will all affect the result that the historical risk event corresponding to the historical risk event node is judged as an abnormal business with risks.

[0126] Therefore, in order to more accurately determine the risk level of the target business and make a more accurate judgment on whether the target business is a risky business, when the server cannot determine that the target business is an abnormal business based on one round of dialogue, the dialogue process continues and the complaint result of the target business is determined through two or more rounds of dialogue.

[0127] The server continues to determine the output text to be sent to the user based on the multiple risk factor nodes corresponding to the at least one historical risk event node, and continues to guide the user to input more business information with richer content to determine more risk factor nodes matching the target business.

[0128] When the target element node has a small impact on at least one determined historical risk event node, it cannot be determined that the target business is a risky business. However, in the input text returned by the user again, because it contains richer business information, it may match the risk element node that has a greater impact on at least one determined historical risk event node, thereby more accurately determining whether the target business has a corresponding relationship with any historical risk event in the risk control map.

[0129] Figure 3 The specific implementation flow chart of step S110 of the complaint handling method provided in this specification is as follows: Figure 3 As shown, the implementation steps of the above step S110 may also include steps S200 to S204.

[0130] S200: Continue to determine an output text to be sent to the user according to the risk factor nodes connected to the at least one historical risk event node.

[0131] S202: Acquire the input text returned by the user to the re-determined output text, and update the risk level according to the returned input text.

[0132] S204: Continue to determine whether the target business can be determined as an abnormal business based on the updated risk level.

[0133] There are multiple implementation methods for the steps S200 to S204.

[0134] In one or more embodiments of the present specification, the output text sent to the user again can be sent once. The specific implementation of the above steps S200 to S204 is as follows:

[0135] In S200, the server generates question texts corresponding to each risk factor node based on multiple risk factor nodes corresponding to at least one historical risk event node, and sends each question text as an output text to the user at one time, guiding the user to input business information related to the risk factor node corresponding to each question text.

[0136] In S204, after the server obtains the input text returned by the user, the server encodes the input text returned again through the word embedding model mentioned in S102, and determines the text features of the input text returned again. Among the element features of each risk element node in the predetermined risk control map, the element features of all risk element nodes connected to at least one risk event node corresponding to the target element node are determined as candidate features. The text features of the input text returned again are matched with each candidate feature, and the first similarity between each candidate feature and the text feature is determined respectively.

[0137] The server determines, for each candidate feature, whether the first similarity corresponding to the candidate feature is greater than a preset first threshold, and if so, determines that the risk factor node corresponding to the candidate feature matches the target business.

[0138] If there is one historical risk event node corresponding to the target element node, the server determines the risk element node corresponding to the historical risk event node that matches the target business, and determines the weights of the edges between these risk element nodes that match the target element and the historical risk event node. The determined weights are added to the risk level determined in step S104 to update the risk level. Then, steps S106 to S110 are re-executed to determine the appeal result of the target business.

[0139] If there are multiple historical risk event nodes corresponding to the target element node, the server determines the risk element node that matches the target business corresponding to each historical risk event node corresponding to the target element node, and determines the weights of the edges between these risk element nodes that match the target element and the historical risk event node. The determined weight is added to the risk level of the historical risk event node determined in step S104 to update the risk level of the historical risk event node.

[0140] In S204, the risk level of each updated historical risk event node is determined, and steps S106 to S110 are repeatedly executed. As long as the risk level of any historical risk event node is greater than a preset risk threshold, the target business can be determined to be an abnormal business.

[0141] In one or more embodiments of the present specification, the server executes the appeal process in the form of multiple rounds of dialogue. The server sequentially accesses the risk factor nodes corresponding to the at least one historical risk event node, and the output text of each round of dialogue is determined only based on the risk factor node currently accessed, that is, each round of dialogue determines whether a risk factor node matches the target business. This allows the server to end the appeal process as soon as possible when it is able to determine that the target business is an abnormal business, without having to access all the risk factor nodes corresponding to the at least one historical risk event node, thereby speeding up the execution efficiency of the appeal process.

[0142] In this embodiment, the output text of the first round of dialogue is preset, and the output text of each subsequent round of dialogue is generated according to the risk factor node corresponding to the common historical risk event node of the target factor node. One risk factor node can determine the output text of a round of dialogue.

[0143] Figure 4 A method provided in the embodiments of this specification Figure 3 The specific implementation flow chart of step S200 is shown as follows: Figure 4 As shown, the above step S200 may specifically include S2001 to S2005. There are two ways of determining the risk level in step S104. Different determination ideas correspond to different specific steps in this embodiment. For the convenience of description, S2001 to S2005 are first described by taking the sequential determination ideas as an example.

[0144] S2001: Determine whether there are any risk factor nodes that have not been visited among the risk factor nodes connected to the at least one historical risk event node. If yes, execute S2003. If no, execute S2005.

[0145] A risk factor node only needs to be visited once, and the output text of a round of dialogue is determined based on the risk factor node, and the user is guided to enter the business information related to the risk factor node in the input text returned by the round of dialogue.

[0146] In the sequential determination approach, the access is performed in units of a historical risk event, and "not visited" refers to the currently visited historical risk event node. Therefore, the server determines whether there are any risk factor nodes that have not been visited in the risk factor nodes connected to the target event node.

[0147] S2003: Determine a pending element node among the risk element nodes that have not been visited and are connected to the at least one historical risk event node; and continue to determine an output text to be sent to the user based on the pending element node.

[0148] When the risk factor nodes connected to the target event node have not been visited, the server determines a pending factor node among the risk factor nodes connected to at least one historical risk event node corresponding to the target factor node. The pending factor node is the risk factor node currently visited by the server.

[0149] This specification does not restrict the method of determining the pending element node, that is, the order of visiting the risk element nodes. The server can randomly determine the pending element node, that is, randomly select a risk element node that has not been visited from each risk element node connected to the target event node as the pending element node.

[0150] Alternatively, in order to obtain the complaint result more quickly and further improve the efficiency of complaint processing, the server can determine the pending element node according to the edge weight. Because the risk level needs to be cumulatively updated according to the edge weight between the pending element node and the target event node, if the target business has risks, during the cumulative update process, if the accumulated weight values ​​are arranged in order from large to small, the risk level will exceed the risk threshold more quickly.

[0151] Specifically, the server uses the risk factor node corresponding to the edge with the largest weight among the risk factor nodes that have not been visited and are connected to the target event node as the pending factor node.

[0152] Then, the server determines the output text to be sent to the user based on the pending element node, and guides the user to return the business information related to the pending element node.

[0153] S2005: Determine whether all risk factor nodes connected to the at least one historical risk event node have been visited.

[0154] When all the risk factor nodes connected to the target event node have been visited, it means that the risk identification based on the target event node is completed.

[0155] The server continues to determine whether there are any historical risk event nodes that have not been visited in the historical risk event nodes corresponding to the target element node.

[0156] If so, jump back to step S104, continue to redefine the target event node among the historical risk event nodes that have not been visited, and loop through steps S104 to S110 through the redetermined target event node to further determine the risk level for the redetermined target event node, and then determine the appeal result of the target business.

[0157] It should be noted that when jumping back to step S104, it means that the server needs to access the next historical risk event node, that is, it needs to re-determine the target event node, because the historical risk event node visited has changed, and the risk level should also be recalculated from zero.

[0158] If not, it means that the dialogue process has reached the maximum round and no longer sends output text to the user. If at this time, the server still cannot determine that the target service is an abnormal service, it determines that the target service is a normal service and removes the limit on the target service.

[0159] Figure 5 A method provided in the embodiments of this specification Figure 3 The specific implementation flow chart of step S202 is shown as follows: Figure 5 As shown, the above step S202 may specifically include S2021 to S2027.

[0160] In this embodiment, S2021 to S2027 are described in accordance with the idea of ​​sequential determination.

[0161] S2021: Obtain the input text returned by the user to the re-determined output text, determine the text features of the input text returned again, match the text features of the input text returned again with the element features of the risk element corresponding to the undetermined element node, and determine a first similarity.

[0162] After the server obtains the input text returned by the user, it encodes the returned input text through the word embedding model mentioned in S102, determines the text features of the returned input text, matches the text features with the feature features of the risk factor corresponding to the undetermined feature node, and determines the first similarity.

[0163] S2023: Determine whether the first similarity is greater than a preset first threshold. If yes, execute S2025. If no, execute S2027.

[0164] S2025: Determine that the target business matches the pending element node, and cumulatively update the risk level through the weight of the edge between the at least one historical event node and the pending element node.

[0165] S2027: Determine that the target business does not match the pending element node, and do not update the risk level.

[0166] The target business matches the pending element node, indicating that the business information of the target business contains the risk factor corresponding to the pending element node, so the risk level of the target event node is cumulatively updated through the weight of the edge between the pending element node and at least one historical event node corresponding to the target element node. The target business does not match the pending element node, indicating that the business information of the target business does not contain the risk factor corresponding to the pending element node, so the weight of the edge between the pending element node and at least one historical event node corresponding to the target element node does not participate in the cumulative update of the risk level of the target event node.

[0167] Still Figure 2 For example, a circle containing an "×" represents a risk factor node that does not match the target business. If the historical risk event node currently accessed by the server is "a", it is necessary to first accumulate and determine the risk level for "a". When the three risk factor nodes corresponding to "a" are traversed, the risk level corresponding to "a" is determined to be w1+w3. Because "2" does not match the target business, w2 does not participate in the accumulation. If w1+w3 is not greater than the risk threshold, the server will redetermine a target event node from "b" and "c". At this time, the server chooses to switch back to step S104. Assuming that the redetermined target event node is "b", the server re-determines the risk level corresponding to "b" as w4 in step S104.

[0168] In S204, the server continues to follow the sequentially determined ideas, and after updating the risk level, it jumps to execute the above step S104, and cyclically executes steps S104 to S110 in each round of dialogue.

[0169] Next, S200 to S204 of this embodiment are described by taking the idea of ​​simultaneous determination as an example. Figure 4 As shown, the above step S200 may specifically include S2001 to S2005.

[0170] In S2001, when the server accesses the risk factor nodes, it does not distinguish the historical risk event nodes to which each risk factor node belongs, and visits the risk factor nodes connected to all the historical event nodes corresponding to the target factor node as a whole. Therefore, the "not visited" in the idea determined at the same time refers to all the risk factor nodes connected to all the historical event nodes corresponding to the target factor node.

[0171] Therefore, in this way of thinking, a risk factor node will only be visited once. That is, for risk factor nodes corresponding to different historical risk event nodes, corresponding to several historical risk event nodes, when visiting this risk factor node, it is necessary to determine the risk levels of several historical risk events.

[0172] In S2003, the server selects a risk factor node that has not been visited from the risk factor nodes connected to all the historical event nodes corresponding to the target event node as a pending factor node. Because in this embodiment, the server regards all the risk factor nodes connected to all the historical event nodes corresponding to the target factor node as a whole, the target event node is no longer determined.

[0173] Similarly, the determination method of the pending element nodes here is similar to the determination method of the pending element nodes in the above sequential determination idea, and can also be determined randomly or by edge weight. Because the historical risk event nodes are not distinguished in this embodiment, the range of the selected pending element nodes is the risk event nodes that have not been visited among the risk element nodes connected to all the historical event nodes corresponding to the target event node.

[0174] In S2005, because the historical risk event nodes are not distinguished in this embodiment, when the risk factor nodes connected to at least one historical risk event node corresponding to the target factor node are accessed, it means that the dialogue process has reached the maximum round and no longer sends output text to the user. If at this time, the server still cannot determine that the target business is an abnormal business, it determines that the target business is a normal business and removes the limit on the target business.

[0175] Next, the S202 of this embodiment will be described by taking the idea of ​​simultaneous determination as an example. Figure 5 As shown, the above step S202 may specifically include S2021 to S2027.

[0176] In the idea of ​​simultaneous determination, the specific implementation steps of S2021 to S2023 are the same as the implementation method in the above-mentioned idea of ​​sequential determination. Please refer to the description of the above-mentioned corresponding content, and no further details will be given here.

[0177] In S2025, if the target business matches the pending element node, the historical risk events that the pending element node and the target element node commonly correspond to are determined, and the risk levels of the determined commonly corresponding historical risk events are cumulatively updated.

[0178] It should be noted that in step S104, the risk level of at least one historical risk event corresponding to the target element node has been determined once. When updating the risk level, the risk levels of all historical risk event nodes corresponding to the target element node may not be updated. The risk levels of the historical risk event nodes corresponding to both the pending element node and the target element node will be updated in this round of dialogue. The historical risk event nodes corresponding to both are historical risk event nodes that are connected to both the target element node and the pending element node.

[0179] In S2027, if the target business does not match the pending element node, the risk level will not be updated during this round of dialogue.

[0180] exist Figure 2 In the example, if the pending element node currently determined by the server is "4", the historical risk event nodes corresponding to the pending element node and the target element node are "b" and "c". The server adds w5 to the risk level corresponding to "b" and adds w8 to the risk level corresponding to "c".

[0181] In the idea of ​​simultaneous determination, the specific implementation steps of S204 are the same as the implementation method in the above-mentioned idea of ​​sequential determination. Please refer to the description of the above-mentioned corresponding contents, and no further details will be given here.

[0182] For the idea of ​​determining in sequence, if the access order of the historical risk event node with risk is relatively high, the number of updates of the risk level will be less, and the server can end the appeal process as soon as possible and determine the appeal result. Otherwise, in extreme cases, if the historical risk event node with risk is the last historical risk event node visited, the event will take a long time. In addition, when there are risk element nodes corresponding to multiple historical risk event nodes in the risk element nodes connected to at least one historical risk event node corresponding to the target element node, the risk element nodes corresponding to the multiple historical risk event nodes will be visited multiple times. In short, in this idea, the order of accessing the historical risk event nodes has a greater impact on the execution rate of the appeal process.

[0183] For the simultaneous determination approach, when the number of historical risk event nodes corresponding to the target element node is large, the number of risk element nodes corresponding to multiple historical risk event nodes will also be large, because when accessing a risk element node, the risk level of each historical risk event node corresponding to the risk element node and the target element node must be updated. The risk level of multiple historical risk event nodes must be updated in one visit, but the advantage is that only one risk level update is performed for a risk element node. In this approach, the number of historical risk nodes corresponding to the target element node has a greater impact on the execution rate of the appeal process.

[0184] It can be seen that both ideas have their own advantages and disadvantages. As for which idea to adopt in the complaint handling process in practice, it needs to be determined based on the characteristics of the business.

[0185] When the server executes the appeal process in the form of multiple rounds of dialogue, if the weights of the edges connected to each historical risk event node have a unified scale, that is, the sum of the weights of the edges connected to different historical event nodes in the risk control map is the same value. In order to speed up the appeal processing process, the server can also set a security threshold, and jointly determine whether the target business matches at least one historical risk event node based on the security threshold and risk threshold. The sum of the security threshold and the risk threshold is equal to the sum of the weights of the edges connected to a historical risk event node.

[0186] Because in this embodiment, the risk level is obtained by summing the weights of the edges connected to each historical risk event node. Therefore, for each historical risk event node, among the risk factor nodes connected to the historical risk event node, if it can be determined that the weight accumulation value of the edge corresponding to the unmatched risk factor node exceeds the safety threshold, then even if all the risk factor nodes connected to the historical risk event node have not been traversed, the weight accumulation value of the edge corresponding to the matching risk factor node can be inferred, that is, the risk level corresponding to the historical risk event node can no longer be greater than the safety threshold. At this time, the risk judgment of the historical risk event node can be terminated in advance, and it is determined that the target business does not match the historical risk event node.

[0187] For example, if the sum of the weights of the edges connected to a risk event node is 1, the risk threshold is 0.6, and the cumulative weight value of the edges corresponding to the currently determined unmatched risk factor nodes is 0.5, it can be inferred that the risk level cannot exceed 0.6.

[0188] Specifically, when the first similarity is not greater than the first threshold, it is determined that the currently visited risk factor node, that is, the pending factor node, does not match the target business. The risk level is not updated, but the safety level is cumulatively updated through the weight of the edge between the pending factor node and at least one historical risk event node.

[0189] Similar to the risk level, the safety level corresponding to different historical risk event nodes should also be calculated separately, and the initial value of the safety level is zero.

[0190] Then, the server determines whether the security level is not less than a preset security threshold. If so, it is determined that the target business does not match at least one historical risk event node. If not, it cannot be determined that the target business does not match at least one historical risk event node, and steps S104 to S110 are still executed.

[0191] In one or more embodiments of the present specification, if the server determines the risk level of each historical risk event node in sequence, it will also determine the safety level of each historical risk event node in sequence accordingly. The safety threshold determined by the server is the safety threshold corresponding to the currently visited risk event node, that is, the target event node.

[0192] When the server determines that the security level is not less than the preset security threshold, it is determined that the target business does not match the target event node. At this time, if the target event node is still connected to a risk factor node that has not been visited, the pending factor node is no longer determined based on the risk factor node that has not been visited connected to the target event node. Instead, it is directly determined whether the target factor node is still connected to a historical risk event node that has not been visited. If so, a new target event node is determined in the at least one historical risk event node that has not been visited corresponding to the target factor node. If not, it is determined that at least one historical risk event node corresponding to the target factor node has been visited.

[0193] In one or more embodiments of the present specification, if the server determines the risk level of each historical risk event node at the same time, it will also determine the safety level of each historical risk event node at the same time. The safety threshold determined by the server is the safety threshold corresponding to each historical risk event node. If after the safety level has been updated according to the user's input text in the current round of dialogue, there is a safety level that exceeds the safety threshold among the safety levels corresponding to each historical risk event node, then the server has been able to determine that the historical risk event node corresponding to the safety level that exceeds the safety threshold does not match the target business, and the cumulative update of the safety level of the historical risk event node can be terminated. In subsequent dialogues, it is only necessary to determine the relationship between the remaining historical risk event nodes and the target business.

[0194] In the above step S102, it is mentioned that the number of target element nodes can be multiple. If the server determines the appeal result through two rounds of dialogue, in S200, the server determines the risk event nodes corresponding to the multiple target element nodes, and then determines the risk element nodes connected to the multiple risk event nodes, and generates the question text corresponding to each risk element node, and uses each question text as the output text.

[0195] If the server determines the appeal result through multiple rounds of dialogue, the execution steps corresponding to determining the ideas sequentially and determining the ideas simultaneously are different.

[0196] According to the sequential determination idea, the server can execute the above steps S102 to S110 for each target element node in turn. If it can be determined that the target business is at risk based on the risk element node connected to at least one historical risk event node corresponding to the target element node, the server can terminate the access to the remaining target element nodes and determine that the target business is an abnormal business. If it is still not possible to determine that the target business is an abnormal business after traversing the risk element nodes connected to at least one historical risk event node corresponding to the target element node, continue to visit the next unvisited target element node and continue to execute the above steps S102 to S110 until all the risk element nodes determined based on all the target element nodes have been visited. If it is still not possible to determine that the target business is an abnormal business, determine that the target business is a normal business.

[0197] Figure 6 A partial schematic diagram of another risk control map provided in an embodiment of this specification, Figure 6 The meaning of the symbols in Figure 2 The same, no further explanation is given here. Figure 6 In the example, the server determines two target element nodes, namely "3" and "6". Then the server determines the risk levels for the historical risk event node corresponding to "3" and the historical risk event node corresponding to "6" according to the above steps S102 to S110.

[0198] Of course, in order to speed up the appeal process, the server may sort the target element nodes in descending order of the second similarities corresponding to the target element nodes, and access the target element nodes in order.

[0199] According to the simultaneous determination idea, the server determines the historical risk event nodes corresponding to the multiple target element nodes, and then determines the risk element nodes connected to the multiple historical risk event nodes as all the risk element nodes to be traversed. The above steps S102 to S110 are performed on the determined risk element nodes to determine the risk level corresponding to each historical risk event node to determine whether the target business is an abnormal business.

[0200] Still Figure 6 For example, when the server simultaneously determines the risk level of each historical risk event node, the risk factor node connected to the historical risk event node corresponding to "3" and the risk factor node connected to the historical risk event node corresponding to "6" are taken as the risk factor nodes to be accessed.

[0201] The above is a complaint handling method provided in one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding complaint handling device, such as Figure 7 shown.

[0202] Figure 7 The schematic diagram of the complaint handling device provided for this manual specifically includes:

[0203] A response module 300 is used to respond to a user's complaint request for a target service by sending a preset output text to the user and obtaining an input text returned by the user to the output text;

[0204] The target element node determination module 302 is used to determine the text features of the input text, match the text features with the element features of each risk element node included in the risk control map, and determine the target element node; the risk control map is composed of historical risk event nodes and risk element nodes, and edges connecting historical risk event nodes and risk element nodes, and the weight of the edge represents the influence of the risk element on the historical risk event;

[0205] A risk level determination module 304 is used to determine at least one historical risk event node corresponding to the target element node, and determine the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node;

[0206] A risk identification module 306 is used to determine whether the target business can be determined as an abnormal business according to the risk level;

[0207] A normal service determination module 308, configured to determine that the target service is an abnormal service and not remove the restriction on the target service;

[0208] The abnormal service determination module 310 is used to determine that the target service is a normal service and, if not, remove restrictions on the target service.

[0209] Optionally, the device further includes a risk control map determination module 312;

[0210] The risk control graph determination module 312 is used to obtain description texts of multiple abnormal historical businesses, and extract events from each description text to determine event keywords and event elements of the multiple abnormal historical businesses.

[0211] According to the determined event keywords, each abnormal historical business is clustered. For each cluster, according to the event elements of each abnormal historical business contained in the cluster, the event elements of the cluster are deduplicated. According to the event keywords of each abnormal historical business contained in the cluster, the characterization keywords of the cluster are determined. According to the characterization keywords of each cluster, the historical risk event nodes corresponding to each cluster are determined, and according to each event element after deduplication of each cluster, the risk factor nodes corresponding to each historical risk event node are determined. For each risk factor node corresponding to each historical risk event node, the ratio of the number of occurrences of the event element corresponding to the risk factor node in each abnormal historical business contained in the cluster corresponding to the historical risk event node to the total number of occurrences of each abnormal historical business is determined. According to the ratio, the weight of the edge between the risk factor node and the historical risk event node is determined. For each historical risk event node, the weights of each edge connecting the historical risk event node are normalized to obtain the risk control map.

[0212] Optionally, the abnormal business determination module 310 is specifically used to continue to determine the output text sent to the user based on the risk factor nodes connected to the at least one historical risk event node when the target business cannot be determined as an abnormal business based on the risk level, obtain the input text returned by the user to the re-determined output text, update the risk level based on the input text returned again, and continue to determine whether the target business can be determined as an abnormal business based on the updated risk level.

[0213] Optionally, the abnormal business determination module 310 is specifically used to determine whether there are any unvisited risk factor nodes among the risk factor nodes connected to the at least one historical risk event node. If so, determine the pending factor nodes among the unvisited risk factor nodes connected to the at least one historical risk event node; based on the pending factor nodes, continue to determine the output text to be sent to the user; if not, determine that the risk factor nodes connected to the at least one historical risk event node have been visited.

[0214] Optionally, the abnormal business determination module 310 is specifically used to determine a target event node from at least one historical risk event that has not been visited and corresponds to the target element node based on the weight of the edge between the at least one historical risk event and the target element node, and to update the risk level of the target business based on the weight of the edge between the target element node and the target event node.

[0215] Optionally, the abnormal business determination module 310 is specifically configured to determine, according to the weights of the edges corresponding to the at least one historical risk event node, a pending element node among the unvisited risk element nodes connected to the at least one historical risk event node.

[0216] Optionally, the abnormal business determination module 310 is specifically used to determine the text features of the input text returned again, match the text features of the input text returned again with the element features of the risk element corresponding to the pending element node, determine a first similarity, and judge whether the first similarity is greater than a preset first threshold. If so, it is determined that the target business matches the pending element node, and the risk level is cumulatively updated through the weight of the edge between the at least one historical event node and the pending element node. If not, it is determined that the target business does not match the pending element node, and the risk level is not updated.

[0217] This specification also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 Provides methods for handling complaints.

[0218] This manual also provides Figure 8 The schematic structure diagram of the electronic device shown in FIG. Figure 8 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Of course, in addition to the software implementation, this specification does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0219] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0220] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0221] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0222] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0223] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0224] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 produce 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.

[0225] 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.

[0226] 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.

[0227] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0228] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0229] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0230] 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.

[0231] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0232] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0233] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0234] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A method for handling complaints, the method comprising: In response to a user's complaint request for a target service, a preset output text is sent to the user, and an input text returned by the user to the output text is obtained; Determine the text features of the input text, match the text features with the feature features of each risk factor node included in the risk control map, and determine the target factor node; the risk control map is composed of historical risk event nodes and risk factor nodes, and edges connecting the historical risk event nodes and the risk factor nodes, and the weight of the edge represents the influence of the risk factor on the historical risk event; Determine at least one historical risk event node corresponding to the target element node, and determine the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node; Determining whether the target business can be determined as an abnormal business according to the risk level; If yes, determine that the target service is an abnormal service, and do not unblock the target service; If not, it is determined that the target business is a normal business, and the target business is unrestricted.

2. The method according to claim 1, wherein the risk control map is determined by the following method, wherein: Acquire description texts of multiple abnormal historical services, and extract events from each description text to determine event keywords and event elements of the multiple abnormal historical services; Cluster each abnormal historical business according to the determined event keywords; For each cluster, according to the event elements of each abnormal historical service contained in the cluster, the event elements of the cluster are deduplicated, and according to the event keywords of each abnormal historical service contained in the cluster, the representation keywords of the cluster are determined; According to the representative keywords of each cluster, determine the historical risk event nodes corresponding to each cluster, and according to the event elements after deduplication of each cluster, determine the risk factor nodes corresponding to each historical risk event node; For each risk factor node corresponding to each historical risk event node, determine the ratio of the number of occurrences of the event element corresponding to the risk factor node in each abnormal historical business included in the cluster corresponding to the historical risk event node to the total number of occurrences in each abnormal historical business, and determine the weight of the edge between the risk factor node and the historical risk event node according to the ratio; For each historical risk event node, the weights of the edges connecting the historical risk event node are normalized to obtain the risk control graph.

3. The method of claim 1, further comprising: When the target business cannot be determined as an abnormal business according to the risk level, continuing to determine an output text to be sent to the user according to the risk factor nodes connected to the at least one historical risk event node; Acquire the input text returned by the user to the re-determined output text, and update the risk level according to the returned input text; It is further determined whether the target business can be determined as an abnormal business based on the updated risk level.

4. The method according to claim 3, further determining the output text to be sent to the user according to the risk factor node connected to the at least one historical risk event node, specifically comprising: Determining whether there are any risk factor nodes that have not been visited among the risk factor nodes connected to the at least one historical risk event node; If so, determining a pending element node among the unvisited risk element nodes connected to the at least one historical risk event node; According to the undetermined element node, continue to determine the output text to be sent to the user; If not, it is determined that all risk factor nodes connected to the at least one historical risk event node have been visited.

5. The method according to claim 3, determining the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node, specifically comprising: Determine a target event node from at least one historical risk event that has not been visited and corresponds to the target element node according to the weight of the edge between the at least one historical risk event and the target element node; The risk level of the target business is updated according to the weight of the edge between the target element node and the target event node.

6. The method according to claim 4 or 5, wherein determining the undetermined element node among the unvisited risk element nodes connected to the at least one historical risk event node specifically comprises: According to the weights of the edges corresponding to the at least one historical risk event node, a pending element node is determined from among the unvisited risk element nodes connected to the at least one historical risk event node.

7. The method according to claim 4, updating the risk level according to the input text returned again, specifically comprising: Determine the text features of the input text returned again, match the text features of the input text returned again with the element features of the risk element corresponding to the undetermined element node, and determine a first similarity; Determining whether the first similarity is greater than a preset first threshold; If so, determine that the target business matches the undetermined element node, and cumulatively update the risk level through the weight of the edge between the at least one historical event node and the undetermined element node; If not, it is determined that the target business does not match the undetermined element node, and the risk level is not updated.

8. A complaint handling device, comprising: A response module, in response to a user's complaint request for a target service, sends a preset output text to the user, and obtains an input text returned by the user to the output text; a target element node determination module, which determines the text features of the input text, matches the text features with the element features of each risk element node included in the risk control map, and determines the target element node; the risk control map is composed of historical risk event nodes and risk element nodes, and edges connecting historical risk event nodes and risk element nodes, and the weight of the edge represents the influence of the risk element on the historical risk event; a risk level determination module, which determines at least one historical risk event node corresponding to the target element node, and determines the risk level of the target business according to the weight of the edge between the at least one historical risk event node and the target element node; A risk identification module, which determines whether the target business can be determined as an abnormal business according to the risk level; A normal business determination module determines that the target business is an abnormal business and does not remove the restriction on the target business. The abnormal business determination module determines that the target business is a normal business, if not, and removes the restriction on the target business.

9. A computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

Images