A system for evaluating the authenticity and accuracy of content generated by large models

Abstract

The present invention provides a system for evaluating the authenticity and accuracy of content generated by a large model, which relates to the field of electronic digital data processing and comprises a content input module, a reference data module, an authenticity evaluation module and an accuracy evaluation module; the content input module is used to receive content information that needs to be evaluated, the reference data module is used to store data information with authenticity, the authenticity evaluation module is used to evaluate the authenticity of the generated content, and the accuracy evaluation module is used to evaluate the degree to which the generated content meets the requirements; the system can obtain data for reflecting the quality of the generated content by disassembling the generated content and comparing them item by item, and processing them in a quantitative manner, so as to facilitate a more reasonable application of the generated content.

Description

Technical Field

[0001] The invention relates to the field of electronic digital data processing, and in particular to a system for evaluating the authenticity and accuracy of content generated by a large model. Background Art

[0002] With the rapid development of artificial intelligence technology, large models based on deep learning have made significant breakthroughs in natural language processing, image generation, speech recognition and other fields. By training massive data, large models can generate logical, coherent and creative content, and are widely used in text generation, automatic question answering, language translation and other scenarios. However, since the basis for large models to generate content mainly comes from training data, they do not have a deep understanding of facts and the ability to obtain real-time information. Therefore, the generated content may contain factual errors, logical inconsistencies or information bias. Therefore, a system is needed to quantitatively evaluate the generated content to facilitate more reasonable application of the generated content in the future.

[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] Now many content evaluation systems have been developed. After a lot of searching and reference, it is found that the existing content evaluation systems are such as the system disclosed in the publication number CN117313722B. These system methods generally include: according to the large language model, the large language model reasoning accuracy index is obtained; according to the large language model and the constructed large language model reasoning accuracy prediction model, the large language model reasoning accuracy prediction index is obtained; according to the reasoning accuracy index and the reasoning accuracy prediction index, the reasoning accuracy prediction result of the large language model is obtained. However, this system is only used to predict the accuracy result, and cannot find the authenticity problem in the generated content. Under the premise that the authenticity cannot be guaranteed, the accuracy also loses its meaning. Summary of the invention

[0005] The purpose of the present invention is to propose a system for evaluating the authenticity and accuracy of content generated by a large model in response to the existing deficiencies.

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

[0007] A system for evaluating the authenticity and accuracy of content generated by a large model, comprising a content input module, a reference data module, an authenticity evaluation module and an accuracy evaluation module;

[0008] The content input module is used to receive content information that needs to be evaluated, the reference data module is used to store data information with authenticity, the authenticity evaluation module is used to evaluate the authenticity of the generated content, and the accuracy evaluation module is used to evaluate the degree to which the generated content meets the requirements;

[0009] The content input module includes a model docking unit, a content storage unit and a model management unit. The model docking unit is used to dock with the interface of the large model and obtain information. The content storage unit is used to save the received information. The model management unit is used to manage the docked large model information.

[0010] The reference data module includes a basic information storage unit, an external information retrieval unit and a data matching unit, wherein the basic information storage unit is used to store authentic basic content, the external information retrieval unit is used to retrieve external data to supplement the basic content, and the data matching unit is used to determine content that matches the generated content;

[0011] The authenticity evaluation module includes a comparison analysis unit, a comparison statistics unit and a comprehensive evaluation unit, wherein the comparison analysis unit is used to compare and analyze the generated content with the matching content, the comparison statistics unit is used to perform statistical processing on the comparison results, and the comprehensive evaluation unit calculates the authenticity index of the generated content based on the statistical results;

[0012] The accuracy evaluation module includes a requirement parsing unit, a content parsing unit and a scoring processing unit, wherein the requirement parsing unit is used to parse the requirement information, the content parsing unit is used to parse the generated content, and the scoring processing unit performs a scoring calculation on the accuracy based on the parsing result;

[0013] Further, the data matching unit includes a matching control processor, a basic matching processor and a data summary register, wherein the matching control processor is used to control the matching process, the basic matching processor is used to obtain matching information from the basic information storage unit, and the data summary register is used to store all matching information;

[0014] The matching control processor controls the matching process including the following steps:

[0015] S1. Obtain generated content information and divide it into multiple matching items;

[0016] S2, the basic matching processor matches each matching item with the content in the basic information storage unit, and saves the matching content to the data summary register;

[0017] S3, sending the matching items that have not been matched to the content to the external information retrieval unit, and saving the retrieved content as the matching content to the data summary register;

[0018] S4, sending the matching items and the summarized matching contents to the authenticity assessment module;

[0019] Further, the comparison and analysis unit includes a vector processor, a similarity calculation processor and a segmentation processor, wherein the vector processor is used to extract features from the matching item and the matching content to obtain a text vector, the similarity calculation processor is used to calculate the similarity between the matching item and the matching content, and the segmentation processor performs segmentation processing based on the similarity result;

[0020] The similarity calculation processor calculates the similarity S(G, R) between the matching item and the corresponding matching content according to the following formula:

[0021] ;

[0022] Among them, G represents the matching item text, R represents the matching content text, n is the dimension of the text vector, and g i Represents the i-th element value of the G text vector, r i Represents the i-th element value of the R text vector;

[0023] Further, the comprehensive evaluation unit includes a real index calculation processor and an advanced judgment processor, the real index calculation processor is used to calculate the real index of the entire generated content, and the advanced judgment processor determines whether to perform subsequent accuracy evaluation based on the real index and sends a corresponding signal to the accuracy evaluation module;

[0024] The real index calculation processor calculates the real index P of the generated content according to the following formula:

[0025] ;

[0026] Where n1 is the number of matches with true labels, n2 is the number of matches with fuzzy labels, and n3 is the number of matches with false labels. is the weight value of the j-th label;

[0027] When the true index P exceeds the threshold, the accuracy assessment module will perform the assessment process;

[0028] Further, the scoring processing unit includes a comparison information processor and an accuracy index calculation processor, the comparison information processor is used to receive the comparison result and perform statistical processing, and the accuracy index calculation processor is used to calculate the accuracy index of the generated content;

[0029] The accuracy index calculation processor calculates the accuracy index Q of the generated content according to the following formula:

[0030] ;

[0031] Among them, m(i) represents the number of keywords that meet the i-th requirement point, m is the number of requirement points, and M is the number of keywords.

[0032] The beneficial effects achieved by the present invention are:

[0033] This system breaks down the generated content into multiple matching items, and compares and analyzes each matching item with the matching content to obtain quantitative data. During the processing, through vector processing, label assignment and weight processing, it can more accurately reflect the authenticity of the generated content. On the basis of meeting the authenticity standards, the accuracy is analyzed, which can more reasonably reflect the quality of the generated content.

[0034] 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

[0035] Figure 1 It is a schematic diagram of the overall structural framework of the present invention;

[0036] Figure 2 It is a schematic diagram of the content input module of the present invention;

[0037] Figure 3 This is a schematic diagram of the reference data module of the present invention;

[0038] Figure 4 This is a schematic diagram of the authenticity assessment module of the present invention;

[0039] Figure 5 This is a schematic diagram of the accuracy assessment module of the present invention;

[0040] Figure 6 This is a sample test data diagram of the present invention. DETAILED DESCRIPTION

[0041] 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.

[0042] Embodiment 1: This embodiment provides a system for evaluating the authenticity and accuracy of content generated by a large model. Figure 1 , including a content input module, a reference data module, an authenticity assessment module and an accuracy assessment module;

[0043] The content input module is used to receive content information that needs to be evaluated, the reference data module is used to store data information with authenticity, the authenticity evaluation module is used to evaluate the authenticity of the generated content, and the accuracy evaluation module is used to evaluate the degree to which the generated content meets the requirements;

[0044] The content input module includes a model docking unit, a content storage unit and a model management unit. The model docking unit is used to dock with the interface of the large model and obtain information. The content storage unit is used to save the received information. The model management unit is used to manage the docked large model information.

[0045] The reference data module includes a basic information storage unit, an external information retrieval unit and a data matching unit, wherein the basic information storage unit is used to store authentic basic content, the external information retrieval unit is used to retrieve external data to supplement the basic content, and the data matching unit is used to determine content that matches the generated content;

[0046] The authenticity evaluation module includes a comparison analysis unit, a comparison statistics unit and a comprehensive evaluation unit, wherein the comparison analysis unit is used to compare and analyze the generated content with the matching content, the comparison statistics unit is used to perform statistical processing on the comparison results, and the comprehensive evaluation unit calculates the authenticity index of the generated content based on the statistical results;

[0047] The accuracy evaluation module includes a requirement parsing unit, a content parsing unit and a scoring processing unit, wherein the requirement parsing unit is used to parse the requirement information, the content parsing unit is used to parse the generated content, and the scoring processing unit performs a scoring calculation on the accuracy based on the parsing result;

[0048] The data matching unit includes a matching control processor, a basic matching processor and a data summary register, wherein the matching control processor is used to control the matching process, the basic matching processor is used to obtain matching information from the basic information storage unit, and the data summary register is used to store all matching information;

[0049] The matching control processor controls the matching process including the following steps:

[0050] S1. Obtain generated content information and divide it into multiple matching items;

[0051] S2, the basic matching processor matches each matching item with the content in the basic information storage unit, and saves the matching content to the data summary register;

[0052] S3, sending the matching items that have not been matched to the content to the external information retrieval unit, and saving the retrieved content as the matching content to the data summary register;

[0053] S4, sending the matching items and the summarized matching contents to the authenticity assessment module;

[0054] The comparison and analysis unit includes a vector processor, a similarity calculation processor and a segmentation processor, wherein the vector processor is used to extract features from the matching item and the matching content to obtain a text vector, the similarity calculation processor is used to calculate the similarity between the matching item and the matching content, and the segmentation processor performs segmentation processing based on the similarity result;

[0055] The similarity calculation processor calculates the similarity S(G, R) between the matching item and the corresponding matching content according to the following formula:

[0056] ;

[0057] Among them, G represents the matching item text, R represents the matching content text, n is the dimension of the text vector, and g i Represents the i-th element value of the G text vector, r i Represents the i-th element value of the R text vector;

[0058] The comprehensive evaluation unit includes a real index calculation processor and an advanced judgment processor, wherein the real index calculation processor is used to calculate the real index of the entire generated content, and the advanced judgment processor determines whether to perform subsequent accuracy evaluation based on the real index and sends a corresponding signal to the accuracy evaluation module;

[0059] The real index calculation processor calculates the real index P of the generated content according to the following formula:

[0060] ;

[0061] Where n1 is the number of matches with true labels, n2 is the number of matches with fuzzy labels, and n3 is the number of matches with false labels. is the weight value of the j-th label;

[0062] When the true index P exceeds the threshold, the accuracy assessment module will perform the assessment process;

[0063] The scoring processing unit includes a comparison information processor and an accuracy index calculation processor, wherein the comparison information processor is used to receive the comparison result and perform statistical processing, and the accuracy index calculation processor is used to calculate the accuracy index of the generated content;

[0064] The accuracy index calculation processor calculates the accuracy index Q of the generated content according to the following formula:

[0065] ;

[0066] Among them, m(i) represents the number of keywords that meet the i-th requirement point, m is the number of requirement points, and M is the number of keywords.

[0067] Embodiment 2: The present invention includes all the contents of Embodiment 1, and provides a system for evaluating the authenticity and accuracy of content generated by a large model, including a content input module, a reference data module, an authenticity evaluation module, and an accuracy evaluation module;

[0068] The content input module is used to receive content information that needs to be evaluated, the reference data module is used to store data information with authenticity, the authenticity evaluation module is used to evaluate the authenticity of the generated content, and the accuracy evaluation module is used to evaluate the degree to which the generated content meets the requirements;

[0069] Combination Figure 2 The content input module includes a model docking unit, a content storage unit and a model management unit. The model docking unit is used to dock with the interface of the large model and obtain information. The content storage unit is used to save the received information. The model management unit is used to manage the docked large model information.

[0070] Combination Figure 3 , the reference data module includes a basic information storage unit, an external information retrieval unit and a data matching unit, the basic information storage unit is used to store the basic content with authenticity, the external information retrieval unit is used to retrieve external data to supplement the basic content, and the data matching unit is used to determine the content that matches the generated content;

[0071] Combination Figure 4The authenticity evaluation module includes a comparison analysis unit, a comparison statistics unit and a comprehensive evaluation unit. The comparison analysis unit is used to compare and analyze the generated content with the matching content, the comparison statistics unit is used to perform statistical processing on the comparison results, and the comprehensive evaluation unit calculates the authenticity index of the generated content based on the statistical results;

[0072] Combination Figure 5 The accuracy evaluation module includes a requirement parsing unit, a content parsing unit and a scoring processing unit, wherein the requirement parsing unit is used to parse the requirement information, the content parsing unit is used to parse the generated content, and the scoring processing unit calculates the accuracy score based on the parsing result;

[0073] The model docking unit includes an interface connection processor and a data transmission processor, wherein the interface connection processor is used to obtain interface information in the model management unit and establish a connection with the interface, and the data transmission processor is used to obtain information from the connected interface;

[0074] The content storage unit includes an information pairing processor, a generated information register and a required information register, wherein the information pairing processor is used to perform pairing processing on the generated information and the required information, the generated information register is used to store the received generated information, and the required information register is used to store the received required information;

[0075] The model management unit includes a model information register and a model editing processor, wherein the model information register is used to store the model information of the large model, and the model editing processor is used to edit the model information;

[0076] The basic information storage unit includes a basic classification processor and a basic information register, wherein the basic classification processor is used to classify the basic information, and the basic information register stores the basic information based on the classification result;

[0077] The external information retrieval unit includes a target locking processor and a retrieval execution processor, wherein the target locking processor is used to lock the retrieval target in the generated content, and the retrieval execution processor is used to obtain relevant information of the retrieval target in the external network;

[0078] The data matching unit includes a matching control processor, a basic matching processor and a data summary register, wherein the matching control processor is used to control the matching process, the basic matching processor is used to obtain matching information from the basic information storage unit, and the data summary register is used to store all matching information;

[0079] The matching control processor controls the matching process including the following steps:

[0080] S1. Obtain generated content information and divide it into multiple matching items;

[0081] S2, the basic matching processor matches each matching item with the content in the basic information storage unit, and saves the matching content to the data summary register;

[0082] S3, sending the matching items that have not been matched to the content to the external information retrieval unit, and saving the retrieved content as the matching content to the data summary register;

[0083] S4, sending the matching items and the summarized matching contents to the authenticity assessment module;

[0084] The comparison and analysis unit includes a vector processor, a similarity calculation processor and a segmentation processor, wherein the vector processor is used to extract features from the matching item and the matching content to obtain a text vector, the similarity calculation processor is used to calculate the similarity between the matching item and the matching content, and the segmentation processor performs segmentation processing based on the similarity result;

[0085] The similarity calculation processor calculates the similarity S(G, R) between the matching item and the corresponding matching content according to the following formula:

[0086] ;

[0087] Among them, G represents the matching item text, R represents the matching content text, n is the dimension of the text vector, and g i Represents the i-th element value of the G text vector, r i Represents the i-th element value of the R text vector;

[0088] The segmentation processor divides the interval [0, 1] into three intervals, corresponding to three labels: true, fuzzy and false, and assigns the corresponding label to the matching item according to the interval in which the similarity falls;

[0089] The comparison statistics unit includes a label statistics processor, a content statistics processor and a weight calculation processor, wherein the label statistics processor is used to count the number of true, fuzzy and false labels, the content statistics processor counts the text volume of each matching item, and the weight calculation processor calculates the weight value based on the text volume;

[0090] The weight calculation processor calculates the relative weight value k(i) of the i-th matching item based on the following formula:

[0091] ;

[0092] Among them, VT(i) represents the text volume of the i-th matching item;

[0093] The weight calculation processor calculates the weight value of each tag according to the following formula: :

[0094] ;

[0095] Among them, n1 represents the number of matches with true labels, n2 represents the number of matches with fuzzy labels, n3 represents the number of matches with false labels, and k j (i) is the relative weight value of the i-th corresponding label matching item;

[0096] The comprehensive evaluation unit includes a real index calculation processor and an advanced judgment processor, wherein the real index calculation processor is used to calculate the real index of the entire generated content, and the advanced judgment processor determines whether to perform subsequent accuracy evaluation based on the real index and sends a corresponding signal to the accuracy evaluation module;

[0097] The real index calculation processor calculates the real index P of the generated content according to the following formula:

[0098] ;

[0099] When the true index P exceeds the threshold, the accuracy assessment module will perform the assessment process;

[0100] The requirement parsing unit includes a standard rule base and a rule parsing processor, wherein the standard rule base is used to store rule information, and the rule parsing processor processes the requirement text based on the rule information to obtain the requirement point;

[0101] The content analysis unit includes a key information extraction processor and a requirement point comparison processor, wherein the key information extraction processor is used to extract keyword information from the generated content, and the requirement point comparison processor is used to compare the satisfaction relationship between the keyword information and the requirement points;

[0102] The scoring processing unit includes a comparison information processor and an accuracy index calculation processor, wherein the comparison information processor is used to receive the comparison result and perform statistical processing, and the accuracy index calculation processor is used to calculate the accuracy index of the generated content;

[0103] The accuracy index calculation processor calculates the accuracy index Q of the generated content according to the following formula:

[0104] ;

[0105] Where m(i) represents the number of keywords that meet the i-th requirement point, m is the number of requirement points, and M is the number of keywords;

[0106] The i and j appearing in the above text are ordinal numbers used to indicate serial numbers;

[0107] The following is part of the code information of this system:

[0108] import numpy as np

[0109] class AuthenticityEvaluationModule:

[0110] def __init__(self):

[0111] self.similarity_threshold_1 = 0.8 # Similarity threshold 1

[0112] self.similarity_threshold_2 = 0.5 # Similarity threshold 2

[0113] def calculate_similarity(self, generated_vector, reference_vector):

[0114] """Calculate the similarity between the generated content and the reference content, using the cosine similarity formula"""

[0115] similarity = np.dot(generated_vector, reference_vector) / (np.linalg.norm(generated_vector) * np.linalg.norm(reference_vector))

[0116] return similarity

[0117] def classify_segment(self, similarity):

[0118] """Classify paragraphs based on similarity and return labels (true, fuzzy, false)"""

[0119] if similarity >= self.similarity_threshold_1:

[0120] return "true"

[0121] elif self.similarity_threshold_2 <= similarity <self.similarity_threshold_1:

[0122] return "blur"

[0123] else:

[0124] return "false"

[0125] def calculate_segment_weight(self, text_length):

[0126] """Calculate weight based on text paragraph length"""

[0127] return np.log(text_length) # Use logarithm to calculate weight

[0128] def calculate_true_index(self, segment_classifications, segment_weights):

[0129] """Calculate the real index according to the formula in the document"""

[0130] true_weight_sum = sum(w for c, w in zip(segment_classifications, segment_weights) if c == "true")

[0131] fuzzy_weight_sum = sum(w for c, w in zip(segment_classifications, segment_weights) if c == "fuzzy")

[0132] false_weight_sum = sum(w for c, w in zip(segment_classifications, segment_weights) if c == "false")

[0133] total_weight_sum = true_weight_sum + fuzzy_weight_sum + false_weight_sum

[0134] true_index = (true_weight_sum * 1.0 + fuzzy_weight_sum * 0.5 +false_weight_sum * 0) / total_weight_sum

[0135] return true_index

[0136] def evaluate_authenticity(self, generated_content, reference_content):

[0137] """Evaluate the authenticity index of the entire generated content"""

[0138] segment_classifications = []

[0139] segment_weights = []

[0140] # Segment processing, assuming that the segmentation is based on natural paragraphs, here use spaces to split into paragraphs

[0141] generated_segments = generated_content.split('. ')

[0142] reference_segments = reference_content.split('. ')

[0143] for gen_seg, ref_seg in zip(generated_segments, reference_segments):

[0144] # Convert each paragraph into a simple vector representation (an advanced model can be used here, such as BERT embedding)

[0145] generated_vector = np.array([ord(char) for char in gen_seg]) # Simple conversion to ASCII vector

[0146] reference_vector = np.array([ord(char) for char in ref_seg])

[0147] # Calculate similarity

[0148] similarity = self.calculate_similarity(generated_vector,reference_vector)

[0149] #Category paragraphs

[0150] classification = self.classify_segment(similarity)

[0151] segment_classifications.append(classification)

[0152] # Calculate paragraph weight (based on paragraph length)

[0153] weight = self.calculate_segment_weight(len(gen_seg))

[0154] segment_weights.append(weight)

[0155] # Calculate the overall real index

[0156] true_index = self.calculate_true_index(segment_classifications, segment_weights)

[0157] print(f"True Index: {true_index}")

[0158] return true_index;

[0159] Now construct a sample text, including ten matching items. The number of true items in the matching items is used as the sample value of the sample text. Enter 50 sample texts as test content into this system and obtain the true index construction chart, such as Figure 6 shown.

[0160] 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 system for evaluating the authenticity and accuracy of content generated by a large model, characterized in that: It includes a content input module, a reference data module, an authenticity assessment module and an accuracy assessment module; The content input module is used to receive content information that needs to be evaluated, the reference data module is used to store data information with authenticity, the authenticity evaluation module is used to evaluate the authenticity of the generated content, and the accuracy evaluation module is used to evaluate the degree to which the generated content meets the requirements; The content input module includes a model docking unit, a content storage unit and a model management unit. The model docking unit is used to dock with the interface of the large model and obtain information. The content storage unit is used to save the received information. The model management unit is used to manage the docked large model information. The reference data module includes a basic information storage unit, an external information retrieval unit and a data matching unit, wherein the basic information storage unit is used to store authentic basic content, the external information retrieval unit is used to retrieve external data to supplement the basic content, and the data matching unit is used to determine content that matches the generated content; The authenticity evaluation module includes a comparison analysis unit, a comparison statistics unit and a comprehensive evaluation unit, wherein the comparison analysis unit is used to compare and analyze the generated content with the matching content, the comparison statistics unit is used to perform statistical processing on the comparison results, and the comprehensive evaluation unit calculates the authenticity index of the generated content based on the statistical results; The accuracy evaluation module includes a requirement parsing unit, a content parsing unit and a scoring processing unit, wherein the requirement parsing unit is used to parse the requirement information, the content parsing unit is used to parse the generated content, and the scoring processing unit performs a scoring calculation on the accuracy based on the parsing result; the data matching unit includes a matching control processor, a basic matching processor and a data summary register, wherein the matching control processor is used to control the matching process, the basic matching processor is used to obtain matching information from the basic information storage unit, and the data summary register is used to store all matching information; The matching control processor controls the matching process including the following steps: S1. Obtain generated content information and divide it into multiple matching items; S2, the basic matching processor matches each matching item with the content in the basic information storage unit, and saves the matching content to the data summary register; S3, sending the matching items that have not been matched to the content to the external information retrieval unit, and saving the retrieved content as the matching content to the data summary register; S4, sending the matching items and the summarized matching contents to the authenticity assessment module; the comparison and analysis unit includes a vector processor, a similarity calculation processor and a segmentation processor, the vector processor is used to extract features from the matching items and the matching contents to obtain text vectors, the similarity calculation processor is used to calculate the similarity between the matching items and the matching contents, and the segmentation processor performs segmentation processing based on the similarity results; The similarity calculation processor calculates the similarity S(G, R) between the matching item and the corresponding matching content according to the following formula: ； Among them, G represents the matching item text, R represents the matching content text, n is the dimension of the text vector, and g i Represents the i-th element value of the G text vector, r i represents the i-th element value of the R text vector; the comprehensive evaluation unit includes a real index calculation processor and an advanced judgment processor, the real index calculation processor is used to calculate the real index of the entire generated content, and the advanced judgment processor determines whether to perform subsequent accuracy evaluation based on the real index and sends a corresponding signal to the accuracy evaluation module; The real index calculation processor calculates the real index P of the generated content according to the following formula: ； Where n1 is the number of matches with true labels, n2 is the number of matches with fuzzy labels, and n3 is the number of matches with false labels. is the weight value of the j-th label; When the true index P exceeds the threshold, the accuracy evaluation module will perform the evaluation process.

2. A system for evaluating the authenticity and accuracy of content generated by a large model as claimed in claim 1, characterized in that: The scoring processing unit includes a comparison information processor and an accuracy index calculation processor, wherein the comparison information processor is used to receive the comparison result and perform statistical processing, and the accuracy index calculation processor is used to calculate the accuracy index of the generated content; The accuracy index calculation processor calculates the accuracy index Q of the generated content according to the following formula: ； Among them, m(i) represents the number of keywords that meet the i-th requirement point, m is the number of requirement points, and M is the number of keywords.

Images