An error detection and correction system for large model generated content

Abstract

The present invention provides an error detection and correction system for content generated by a large model, which relates to the field of electrical digital data processing, and comprises a content receiving module, a content detection module, a content correction module and a feedback output module; the content receiving module is used to receive the generated content of the large model, the content detection module is used to detect the error information existing in the generated content, the content correction module performs correction processing on the generated content based on the detection result, and the feedback output module is used to summarize the corrected content and feed it back to the corresponding large model; in addition to being able to detect basic error content, the system can also detect and correct the existing contradictory errors according to the context, and optimize the output content of the large model.

Description

Technical Field

[0001] The present invention relates to the field of electrical digital data processing, and in particular to an error detection and correction system for large model generated content. Background Art

[0002] As a cutting-edge technology in the field of artificial intelligence, big models have made significant progress in natural language processing, computer vision and other fields in recent years. Based on deep learning technology, big models can generate high-quality text, images, codes and other content through massive data training, which greatly improves the machine's language understanding and generation capabilities. Such models usually have billions or even tens of billions of parameters and can model complex semantic structures, thereby realizing the automated processing of multiple tasks. However, although the current generated content can avoid most basic errors, it is still prone to inconsistent errors. Such errors are not easy to detect, so a more complete detection and correction system is needed to optimize the generated content.

[0003] The foregoing discussion of the background art is intended only to facilitate an understanding of the present invention. This discussion does not acknowledge or admit that any of the material referred to is part of the common general knowledge.

[0004] Many content detection systems have been developed. After a lot of searching and reference, it is found that the existing content detection systems are such as the system disclosed in the publication number CN103176984B. These system methods generally include: crawling the information in the set user-generated content information source and generating a user-generated content web page set, and then annotating the user-generated content web pages to obtain a candidate deceptive junk opinion set; 2) filtering the samples in the set to form an accurate deceptive junk opinion set; 3) extracting features from the accurate deceptive junk opinion set and the unannotated user-generated content web page set samples from four dimensions: opinion author, opinion content, content distribution, and links, and generating feature vectors for each dimension; 4) using machine learning methods to establish a deceptive junk opinion detection model for each dimension; 5) using the deceptive junk opinion detection model to detect deceptive junk opinions on newly crawled user-generated content web pages. However, this system is only suitable for detecting more obvious erroneous information, and it is not easy to find contradictory errors caused by previous and subsequent connections. Summary of the invention

[0005] The purpose of the present invention is to propose an error detection and correction system for large model generated content in view of the existing deficiencies.

[0006] The present invention adopts the following technical solution:

[0007] A large model generated content error detection and correction system, comprising a content receiving module, a content detection module, a content correction module and a feedback output module;

[0008] The content receiving module is used to receive the generated content of the large model, the content detection module is used to detect the error information in the generated content, the content correction module corrects the generated content based on the detection result, and the feedback output module is used to summarize the corrected content and feed it back to the corresponding large model;

[0009] The content receiving module includes a model management unit, an interface connection unit and an information transmission unit, wherein the model management unit is used to manage the large model information to be connected, the interface connection unit is connected with the output interface of the large model based on the large model information, and the information transmission unit is used to receive and store the generated content output by the large model;

[0010] The content detection module includes an error analysis unit, an attribute evaluation unit, and a multi-layer detection unit, wherein the error analysis unit is used to analyze the basic error of the analysis target, the attribute evaluation unit is used to assign attribute information to the analysis target, and the multi-layer detection unit detects the relational errors between multiple analysis targets based on the attribute information;

[0011] The content correction module includes an error receiving unit, an error matching unit and a correction replacement unit, wherein the error receiving unit is used to receive error information sent by the content detection module, the error matching unit matches a corresponding correction method based on the error information, and the correction replacement unit replaces information on the generated content based on the correction method;

[0012] The feedback output module includes a project management unit, a summary management unit and a feedback processing unit. The project management unit is used to create a detection and correction project and receive corresponding data. The summary management unit is used to summarize the correction content and determine whether the generated content in the project has been processed. The feedback processing unit is used to send the corrected generated content to the large model corresponding to the project.

[0013] Furthermore, the error analysis unit includes a target extraction processor, a basic rule register and an error comparison processor, wherein the target extraction processor is used to extract an analysis target from the generated content, the basic rule register is used to store basic rule information, and the error comparison processor performs comparison processing on the analysis target based on the rule information.

[0014] Furthermore, the multi-layer detection unit includes an attribute chain processor, a conflict assignment processor and an error screening processor. The attribute chain processor is used to screen out multiple analysis targets with attribute association from the analysis targets to form a chain target. The conflict assignment processor is used to perform conflicting assignments on the analysis targets in the chain targets. The error screening processor screens out analysis targets that are errors based on the assignments.

[0015] Furthermore, the process of the error screening processor screening out analysis targets from chain targets includes the following steps:

[0016] S1. Calculate the conflict difference between two adjacent analysis targets And the conflict property Nt:

[0017] ;

[0018] ;

[0019] Among them, v1 and v2 are the conflict values ​​of adjacent analysis targets;

[0020] S2, splitting between two analysis targets with conflict property Nt of -1, and dividing the chain target into multiple target segments;

[0021] S2. Calculate the maximum cumulative conflict value Av of two adjacent target segments according to the following formula:

[0022] ;

[0023] Where n is the starting conflict difference sequence number in the first target segment, m is the ending conflict difference sequence number in the second target segment, w is the segmentation conflict difference sequence number of the two target segments, and P(n, m, w) is the cumulative conflict function;

[0024] S3, filter out the sequence number parameters n, m and w whose maximum cumulative conflict value is greater than the threshold, and use the wnth analysis target and the w+m+1th analysis target in the chain target as error pairing;

[0025] S4. Calculate the correction index of the analysis target in the error pairing according to the following formula;

[0026] S5. Send the analysis target with a larger correction index as an error to the content correction module.

[0027] Furthermore, the calculation formula of the correction index of the analysis target in step S4 is as follows:

[0028] ;

[0029] ;

[0030] Where Q(x) represents the correction index of the x-th analysis target, represents the conflict difference between the x-th analysis target and the i-th analysis target, Q(y) represents the correction index of the y-th analysis target, represents the conflict difference between the i-th analysis target and the y-th analysis target, and x and y represent the sequence numbers of the analysis targets in the error pairing.

[0031] The beneficial effects achieved by the present invention are:

[0032] This system divides the generated content into multiple analysis targets. In addition to being able to process the analysis targets individually to detect common basic errors, it also assigns attributes to the analysis targets, obtains chain targets composed of multiple analysis targets through the attributes, and performs correlation quantitative processing on the analysis targets in the chain targets. It can detect the contradictions before and after and make corrections, which can greatly optimize the generated content of large models.

[0033] To further understand the features and technical contents of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are only for reference and description and are not intended to limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall structural framework of the present invention;

[0035] Figure 2 It is a schematic diagram of the content receiving module of the present invention;

[0036] Figure 3 It is a schematic diagram of the content detection module of the present invention;

[0037] Figure 4 This is a schematic diagram of the error analysis unit of the present invention;

[0038] Figure 5 It is a schematic diagram of the multi-layer detection unit of the present invention;

[0039] Figure 6 This is a comparison chart of the detection effects of the system of the present invention and a common system. DETAILED DESCRIPTION

[0040] The following is an explanation of the embodiments of the present invention through specific embodiments. Those skilled in the art can understand the advantages and effects of the present invention from the contents disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and the details in this specification can also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. In addition, the drawings of the present invention are only simple schematic illustrations and are not depicted according to actual sizes. It is stated in advance. The following embodiments will further explain the relevant technical contents of the present invention in detail, but the disclosed contents are not intended to limit the scope of protection of the present invention.

[0041] Embodiment 1: This system provides an error detection and correction system for large model generated content, combined with Figure 1 , including a content receiving module, a content detecting module, a content correcting module and a feedback output module;

[0042] The content receiving module is used to receive the generated content of the large model, the content detection module is used to detect the error information in the generated content, the content correction module corrects the generated content based on the detection result, and the feedback output module is used to summarize the corrected content and feed it back to the corresponding large model;

[0043] The content receiving module includes a model management unit, an interface connection unit and an information transmission unit, wherein the model management unit is used to manage the large model information to be connected, the interface connection unit is connected with the output interface of the large model based on the large model information, and the information transmission unit is used to receive and store the generated content output by the large model;

[0044] The content detection module includes an error analysis unit, an attribute evaluation unit, and a multi-layer detection unit, wherein the error analysis unit is used to analyze the basic error of the analysis target, the attribute evaluation unit is used to assign attribute information to the analysis target, and the multi-layer detection unit detects the relational errors between multiple analysis targets based on the attribute information;

[0045] The content correction module includes an error receiving unit, an error matching unit and a correction replacement unit, wherein the error receiving unit is used to receive error information sent by the content detection module, the error matching unit matches a corresponding correction method based on the error information, and the correction replacement unit replaces information on the generated content based on the correction method;

[0046] The feedback output module includes a project management unit, a summary management unit and a feedback processing unit. The project management unit is used to create a detection and correction project and receive corresponding data. The summary management unit is used to summarize the correction content and determine whether the generated content in the project has been processed. The feedback processing unit is used to send the corrected generated content to the large model corresponding to the project.

[0047] The error analysis unit includes a target extraction processor, a basic rule register and an error comparison processor. The target extraction processor is used to extract the analysis target from the generated content. The basic rule register is used to store basic rule information. The error comparison processor compares the analysis target based on the rule information.

[0048] The multi-layer detection unit includes an attribute chain processor, a conflict assignment processor and an error screening processor. The attribute chain processor is used to screen out multiple analysis targets with attribute association from the analysis targets to form a chain target. The conflict assignment processor is used to perform conflicting assignments on the analysis targets in the chain targets. The error screening processor screens out analysis targets that are errors based on the assignments.

[0049] The process of the error screening processor screening out analysis targets from chain targets comprises the following steps:

[0050] S1. Calculate the conflict difference between two adjacent analysis targets And the conflict property Nt:

[0051] ;

[0052] ;

[0053] Among them, v1 and v2 are the conflict values ​​of adjacent analysis targets;

[0054] S2, splitting between two analysis targets with conflict property Nt of -1, and dividing the chain target into multiple target segments;

[0055] S2. Calculate the maximum cumulative conflict value Av of two adjacent target segments according to the following formula:

[0056] ;

[0057] Where n is the starting conflict difference sequence number in the first target segment, m is the ending conflict difference sequence number in the second target segment, w is the segmentation conflict difference sequence number of the two target segments, and P(n, m, w) is the cumulative conflict function;

[0058] S3, filter out the sequence number parameters n, m and w whose maximum cumulative conflict value is greater than the threshold, and use the wnth analysis target and the w+m+1th analysis target in the chain target as error pairing;

[0059] S4. Calculate the correction index of the analysis target in the error pairing according to the following formula;

[0060] S5. Send the analysis target with a larger correction index as an error to the content correction module.

[0061] The calculation formula of the correction index of the analysis target in step S4 is as follows:

[0062] ;

[0063] ;

[0064] Where Q(x) represents the correction index of the x-th analysis target, represents the conflict difference between the x-th analysis target and the i-th analysis target, Q(y) represents the correction index of the y-th analysis target, represents the conflict difference between the i-th analysis target and the y-th analysis target, and x and y represent the sequence numbers of the analysis targets in the error pairing.

[0065] Embodiment 2: This embodiment includes all the contents of Embodiment 1, and provides an error detection and correction system for content generated by a large model, including a content receiving module, a content detection module, a content correction module and a feedback output module.

[0066] The content receiving module is used to receive the generated content of the large model, the content detection module is used to detect error information in the generated content, the content correction module corrects the generated content based on the detection result, and the feedback output module is used to summarize the corrected content and feed it back to the corresponding large model.

[0067] Combination Figure 2 The content receiving module includes a model management unit, an interface connection unit and an information transmission unit. The model management unit is used to manage the large model information to be connected. The interface connection unit is connected to the output interface of the large model based on the large model information. The information transmission unit is used to receive and store the generated content output by the large model.

[0068] Combination Figure 3 The content detection module includes an error analysis unit, an attribute evaluation unit and a multi-layer detection unit. The error analysis unit is used to analyze the basic error of the analysis target, the attribute evaluation unit is used to assign attribute information to the analysis target, and the multi-layer detection unit detects the relational errors between multiple analysis targets based on the attribute information.

[0069] The content correction module includes an error receiving unit, an error matching unit and a correction replacement unit. The error receiving unit is used to receive error information sent by the content detection module. The error matching unit matches the corresponding correction method based on the error information. The correction replacement unit replaces the generated content based on the correction method.

[0070] The feedback output module includes a project management unit, a summary management unit and a feedback processing unit. The project management unit is used to create a detection and correction project and receive corresponding data. The summary management unit is used to summarize the correction content and determine whether the generated content in the project has been processed. The feedback processing unit is used to send the corrected generated content to the large model corresponding to the project.

[0071] The model management unit includes a model information register, a model information editing processor and a model information output processor. The model information register is used to store basic information of the large model, the model information editing processor is used to edit the basic information of the large model, and the model information output processor is used to process output applications of model information.

[0072] The interface connection unit includes an information application processor, a model connection processor and a status monitoring processor. The information application processor is used to submit an interface information application to the model management unit and obtain interface information. The model connection processor establishes a connection relationship with the corresponding large model based on the interface information. The status monitoring processor is used to monitor the connection status with the large model.

[0073] The information transmission unit includes a transmission judgment processor and a generated text register. The transmission judgment processor is used to judge whether information needs to be received, and the generated text register is used to store the received generated content text information.

[0074] Combination Figure 4 The error analysis unit includes a target extraction processor, a basic rule register and an error comparison processor. The target extraction processor is used to extract the analysis target from the generated content, the basic rule register is used to store the basic rule information, and the error comparison processor performs comparison processing on the analysis target based on the rule information.

[0075] The attribute evaluation processor includes a target positioning processor, an attribute rule register and an attribute marking processor. The target positioning processor is used to locate the position of the analysis target in the generated content, the attribute rule register is used to store attribute rule information, and the attribute marking processor attributes each analysis target based on the positioning information and the attribute rule.

[0076] Combination Figure 5 The multi-layer detection unit includes an attribute chain processor, a conflict assignment processor and an error screening processor. The attribute chain processor is used to screen out multiple analysis targets with attribute association from the analysis targets to form a chain target. The conflict assignment processor is used to perform conflicting assignments on the analysis targets in the chain target. The error screening processor screens out analysis targets that are errors based on the assignments.

[0077] Attributes include multiple categories, each category includes multiple attribute names, and each analysis target tag may have one or more attribute names. When two analysis targets have attribute names under the same category, the two analysis targets have attribute association.

[0078] The same attribute type contained in all analysis targets in a chain target is called the target category of the chain target.

[0079] Each attribute name has an attribute value under the corresponding attribute category, and the range of the attribute value is [-1, 1]. The conflict assignment processor directly uses the attribute value of the attribute name under the target category included in the analysis target as the conflict value of the analysis target.

[0080] The process of the error screening processor screening out analysis targets from chain targets comprises the following steps:

[0081] S1. Calculate the conflict difference between two adjacent analysis targets And the conflict property Nt:

[0082] ;

[0083] ;

[0084] Among them, v1 and v2 are the conflict values ​​of adjacent analysis targets;

[0085] S2, splitting between two analysis targets with conflict property Nt of -1, and dividing the chain target into multiple target segments;

[0086] S2. Calculate the maximum cumulative conflict value Av of two adjacent target segments according to the following formula:

[0087] ;

[0088] Where n is the starting conflict difference sequence number in the first target segment, m is the ending conflict difference sequence number in the second target segment, w is the segmentation conflict difference sequence number of the two target segments, and P(n, m, w) is the cumulative conflict function;

[0089] The specific expression of the cumulative conflict function is as follows:

[0090] ;

[0091] S3, filter out the sequence number parameters n, m and w whose maximum cumulative conflict value is greater than the threshold, and use the wnth analysis target and the w+m+1th analysis target in the chain target as error pairing. For the convenience of representation and description, x and y are used to represent the sequence numbers of the analysis targets in the error pairing respectively;

[0092] S4. Calculate the correction index of the analysis target in the error pairing according to the following formula:

[0093] ;

[0094] ;

[0095] Where Q(x) represents the correction index of the x-th analysis target, represents the conflict difference between the x-th analysis target and the i-th analysis target, Q(y) represents the correction index of the y-th analysis target, represents the conflict difference between the i-th analysis target and the y-th analysis target;

[0096] S5. Send the analysis target with a larger correction index as an error to the content correction module.

[0097] The error receiving unit includes an error information register and an error management processor. The error information register is used to receive and store error information, and the error management processor is used to manage the stored error information.

[0098] The error matching unit includes a matching relationship register and a matching output processor, wherein the matching relationship register is used to store the matching relationship between the error type and the correction method, and the matching output processor is used to match the error information and send the analysis target and the correction method to the correction replacement unit.

[0099] The analysis target is included in the error information.

[0100] The correction replacement unit includes a replacement selection processor and a replacement feedback processor. The replacement selection processor selects content to be replaced from the analysis target based on the correction method and performs replacement processing. The replacement feedback processor is used to send the replaced analysis target to the feedback output module.

[0101] The i and j appearing in the above text are both ordinal numbers used to indicate serial numbers.

[0102] The following information is part of the code information of this system:

[0103] import numpy as np

[0104] # Define function to calculate conflict property Nt

[0105] def calculate_conflict_type(v1, v2):

[0106] return np.sign(v2 - v1) # returns -1, 0 or 1

[0107] # Define cumulative conflict function P(n, m, w)

[0108] def cumulative_conflict(P, n, m, w):

[0109] return np.sum(P[n:w]) + np.sum(P[w+1:m+1])

[0110] # Define the screening process function

[0111] def error_screening(conflict_values, threshold):

[0112] length = len(conflict_values)

[0113] # S1: Calculate the conflict difference and conflict properties Nt of adjacent analysis targets

[0114] conflict_diffs = [conflict_values[i+1] - conflict_values[i] for iin range(length - 1)]

[0115] conflict_types = [calculate_conflict_type(conflict_values[i],conflict_values[i+1]) for i in range(length - 1)]

[0116] # S2: Split between analysis targets where Nt is -1

[0117] segments = []

[0118] start = 0

[0119] for i in range(1, len(conflict_types)):

[0120] if conflict_types[i] == -1:

[0121] segments.append((start, i)) # Add the start and end positions of the segment

[0122] start = i + 1

[0123] if start < length - 1:

[0124] segments.append((start, length - 1))

[0125] # S3: Calculate the maximum cumulative conflict value of adjacent target segments and filter out the sequence numbers that are greater than the threshold

[0126] max_cumulative_conflict = 0

[0127] n, m, w = 0, 0, 0

[0128] for i in range(len(segments) - 1):

[0129] for j in range(i + 1, len(segments)):

[0130] n_i, m_i = segments[i][0], segments[i][1]

[0131] n_j, m_j = segments[j][0], segments[j][1]

[0132] for w_i in range(n_i, m_j + 1):

[0133] cumulative_conflict_value = cumulative_conflict(conflict_diffs, n_i, m_j, w_i)

[0134] if cumulative_conflict_value > max_cumulative_conflict:

[0135] max_cumulative_conflict = cumulative_conflict_value

[0136] n, m, w = n_i, m_j, w_i

[0137] if max_cumulative_conflict > threshold:

[0138] x = w - n

[0139] y = w + m + 1

[0140] return x, y, max_cumulative_conflict

[0141] return None, None, None

[0142] # S4: Calculate the correction index

[0143] def calculate_correction_index(conflict_diffs, x, y):

[0144] Q_x = np.sum([abs(diff) for diff in conflict_diffs[:x]])

[0145] Q_y = np.sum([abs(diff) for diff in conflict_diffs[y:]])

[0146] return Q_x, Q_y

[0147] # Example data

[0148] conflict_values ​​= [0.1, 0.3, 0.2, -0.5, -0.8, 0.5] # Example conflict values

[0149] threshold = 1.0 # Set the threshold

[0150] # Call the error filtering function

[0151] x, y, max_cumulative_conflict = error_screening(conflict_values,threshold)

[0152] if x is not None and y is not None:

[0153] # Calculate the correction index

[0154] Q_x, Q_y = calculate_correction_index(conflict_values, x, y)

[0155] # S5: Take the analysis target with a larger correction index as the error

[0156] if Q_x > Q_y:

[0157] print(f"Analysis target {x} is error, correction index: {Q_x}")

[0158] else:

[0159] print(f"Analysis target {y} is error, correction index: {Q_y}")

[0160] else:

[0161] print("No error analysis target that meets the conditions was found").

[0162] Now we construct 10 samples, each of which contains 10 errors. We test them in this system and in the common system respectively, and get the following results: Figure 6 The effect diagram shown.

[0163] The contents disclosed above are only preferred feasible embodiments of the present invention, and do not limit the protection scope of the present invention. Therefore, all equivalent technical changes made using the contents of the present invention specification and drawings are included in the protection scope of the present invention. In addition, the elements therein can be updated as technology develops.

Claims

1. A large model generated content error detection and correction system, characterized in that: It includes a content receiving module, a content detecting module, a content correcting module and a feedback output module; The content receiving module is used to receive the generated content of the large model, the content detection module is used to detect the error information in the generated content, the content correction module corrects the generated content based on the detection result, and the feedback output module is used to summarize the corrected content and feed it back to the corresponding large model; The content receiving module includes a model management unit, an interface connection unit and an information transmission unit, wherein the model management unit is used to manage the large model information to be connected, the interface connection unit is connected with the output interface of the large model based on the large model information, and the information transmission unit is used to receive and store the generated content output by the large model; The content detection module includes an error analysis unit, an attribute evaluation unit, and a multi-layer detection unit, wherein the error analysis unit is used to analyze the basic error of the analysis target, the attribute evaluation unit is used to assign attribute information to the analysis target, and the multi-layer detection unit detects the relational errors between multiple analysis targets based on the attribute information; The content correction module includes an error receiving unit, an error matching unit and a correction replacement unit, wherein the error receiving unit is used to receive error information sent by the content detection module, the error matching unit matches a corresponding correction method based on the error information, and the correction replacement unit replaces information on the generated content based on the correction method; The feedback output module includes a project management unit, a summary management unit and a feedback processing unit. The project management unit is used to create a detection and correction project and receive corresponding data. The summary management unit is used to summarize the correction content and determine whether the generated content in the project has been processed. The feedback processing unit is used to send the corrected generated content to the large model corresponding to the project. The multi-layer detection unit includes an attribute chain processor, a conflict assignment processor and an error screening processor, wherein the attribute chain processor is used to screen out multiple analysis targets with attribute association from the analysis targets to form a chain target, the conflict assignment processor is used to perform conflict assignment on the analysis targets in the chain target, and the error screening processor screens out the analysis targets as errors based on the assignment; The process of the error screening processor screening out analysis targets from chain targets comprises the following steps: S1. Calculate the conflict difference between two adjacent analysis targets And the conflict property Nt: ； ； Among them, v1 and v2 are the conflict values ​​of adjacent analysis targets; S2, splitting between two analysis targets with conflict property Nt of -1, and dividing the chain target into multiple target segments; S3. Calculate the maximum cumulative conflict value Av of two adjacent target segments according to the following formula: ； Where n is the starting conflict difference sequence number in the first target segment, m is the ending conflict difference sequence number in the second target segment, w is the segmentation conflict difference sequence number of the two target segments, and P(n, m, w) is the cumulative conflict function; Filter out the serial number parameters n, m and w whose maximum cumulative conflict value is greater than the threshold, and use the wnth analysis target and the w+m+1th analysis target in the chain target as error pairing; S4. Calculate the correction index of the analysis target in the error pairing according to the following formula; S5. Send the analysis target with a larger correction index as an error to the content correction module.

2. The error detection and correction system for large model generated content as claimed in claim 1, characterized in that: The error analysis unit includes a target extraction processor, a basic rule register and an error comparison processor. The target extraction processor is used to extract the analysis target from the generated content. The basic rule register is used to store basic rule information. The error comparison processor compares the analysis target based on the rule information.

3. The error detection and correction system for large model generated content as claimed in claim 2, characterized in that: The calculation formula of the correction index of the analysis target in step S4 is as follows: ； ； Where Q(x) represents the correction index of the x-th analysis target, represents the conflict difference between the x-th analysis target and the i-th analysis target, Q(y) represents the correction index of the y-th analysis target, represents the conflict difference between the ith analysis target and the yth analysis target, and x and y represent the sequence numbers of the analysis targets in the error pairing.

Images