Judgment method and device for AI generated academic text and related product

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a judgment method and device for AI academic text generation and related products. The method comprises the steps of classifying academic texts to be judged to obtain a target subject category; rewriting the academic text to be judged by using the large model to obtain a rewritten text; calculating the target information amount of the academic text to be judged and the average information amount of the rewritten text, and subtracting the target information amount and the average information amount to obtain a target information amount difference value; based on the first difference value distribution condition data, calculating a first AI generation vote number and a first human writing vote number by utilizing the target subject category and the standardized target information amount difference value; calculating a possibility score that the academic text to be judged is an AI generation text based on the two types of votes; and judging whether the academic text to be judged is generated by the AI or not according to a preset threshold value. According to the method, the source of the academic text to be judged can be accurately judged by using one piece of difference value distribution data without independently training a model for each subject category.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and more specifically, to a method, device and related products for determining whether AI generates academic texts. Background Art

[0002] In today's digital and intelligent era, academic research and exchanges are unprecedentedly active. With the rapid development of artificial intelligence technology, its application in academic text generation has become more and more extensive. However, in the academic field, it is crucial to ensure the originality and reliability of research results. If the source of the generation of academic texts cannot be clearly identified, the balance and health of the academic ecology may be destroyed. At the same time, for the academic evaluation system, whether the generation method of the text can be accurately judged is also directly related to the fair assessment of the value of academic achievements. Therefore, exploring the judgment method of AI-generated academic texts has become an urgent and critical task in the current academic development.

[0003] At present, the common method of detecting whether an academic text is generated by AI mainly focuses on using the differences in text features between academic content generated by AI and academic content written by humans, such as differences in language style, logical structure, word usage habits and other features, to automatically classify. However, academic texts in different disciplines are quite different. Therefore, when using existing academic content judgment methods to identify academic texts generated by AI, if you want to achieve a high recognition accuracy in each subject area, you need to train different models or classification methods for different disciplines, which will require more resources and energy. The recognition effect of a single model on academic content in different disciplines also varies greatly. Therefore, there is an urgent need to solve this technical problem. Summary of the invention

[0004] In view of the above situation, the embodiments of the present disclosure provide a method, device and related products for judging whether academic texts are generated by AI, aiming to solve the above problems or at least partially solve the above problems.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for determining whether an AI-generated academic text is generated, the method comprising:

[0006] Classify the acquired academic texts to be judged to obtain the target subject category;

[0007] Rewrite the academic text to be judged by using the first preset large model and the preset rewriting instruction to obtain at least one rewritten text; the preset rewriting instruction includes the following rewriting requirements: maintaining the original meaning and retaining the original style;

[0008] Calculating the target information amount data of the academic text to be judged and the average information amount data of the at least one rewritten text; subtracting the target information amount data from the average information amount data to obtain a target information amount difference;

[0009] Based on the pre-constructed first difference distribution data, the first AI-generated vote number and the first human-written vote number are calculated using the target subject category and the standardized target information difference; the first difference distribution data reflects the distribution of the information difference of the AI-generated sample text after rewriting and the information difference of the human-written sample text after rewriting in different subject categories in different intervals;

[0010] Based on the first AI generated vote number and the first human written vote number, a possibility score of the academic text to be judged being an AI generated text is calculated; and based on the possibility score and a preset threshold, it is judged whether the academic text to be judged is generated by AI to obtain a judgment result.

[0011] In a second aspect, the present disclosure also provides a device for determining whether an AI-generated academic text is generated, the device comprising:

[0012] The classification module is used to classify the acquired academic texts to be judged and obtain the target subject category;

[0013] A rewriting module is used to rewrite the academic text to be judged by using a first preset large model and a preset rewriting instruction to obtain at least one rewritten text; the preset rewriting instruction includes the following rewriting requirements: maintaining the original meaning and retaining the original style;

[0014] A calculation module, used for calculating the target information amount data of the academic text to be judged and the average information amount data of the at least one rewritten text; subtracting the target information amount data from the average information amount data to obtain a target information amount difference;

[0015] A voting module is used to calculate the first AI-generated vote count and the first human-written vote count based on the pre-constructed first difference distribution data, using the target subject category and the standardized target information difference; the first difference distribution data reflects the distribution of the information difference of the AI-generated sample text after rewriting and the information difference of the human-written sample text after rewriting in different subject categories in different intervals;

[0016] The judgment module is used to calculate the possibility score of the academic text to be judged being an AI-generated text based on the first AI-generated vote number and the first human-written vote number; and judge whether the academic text to be judged is generated by AI based on the possibility score and a preset threshold to obtain a judgment result.

[0017] In a third aspect, an embodiment of the present disclosure further provides an electronic device, comprising: a processor; and a memory arranged to store computer executable instructions, which, when executed, cause the processor to execute the steps of the above-mentioned method for determining whether AI generates academic text.

[0018] In a fourth aspect, an embodiment of the present disclosure also provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes the steps of the above-mentioned method for judging whether AI generates academic text.

[0019] By means of the above technical scheme, the judgment method, device and related products of AI-generated academic text provided by the embodiment of the present disclosure can use the pre-constructed first difference distribution data to identify the source of the academic text to be judged. The first difference distribution data is equivalent to a reference standard or feature pattern, which reflects the difference in information before and after rewriting between AI-generated sample texts and human-written sample texts under a large number of different subject categories, and the distribution law in different intervals reflects the characteristic difference in information expression between AI-generated texts and human-written texts. During implementation, according to the target subject category corresponding to the academic text to be judged and the target information difference after standardization, the total number of information difference after rewriting of AI-generated sample texts and the total number of information difference after rewriting of human-written sample texts can be obtained under the standard given by the first difference distribution data, and these two total numbers are respectively used as the first AI-generated vote number and the first human-written vote number for the academic text to be judged. Then, the possibility score of the academic text to be judged as an AI-generated text is calculated based on these votes, and then judged in combination with the preset threshold, thereby achieving the purpose of accurately distinguishing whether the academic text is generated by AI. Compared with the prior art, the technical solution provided by the embodiments of the present disclosure does not require separate model training for each subject category. By using a difference interval distribution data, it can accurately determine whether the academic text to be judged is generated by AI, while also greatly saving time and labor costs.

[0020] The above description is only an overview of the technical solution of the present disclosure. In order to more clearly understand the technical means of the present disclosure, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present disclosure more obvious and easy to understand, the specific implementation methods of the present disclosure are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0022] Figure 1 A schematic diagram of a process for determining an AI-generated academic text provided by an embodiment of the present disclosure is shown;

[0023] Figure 2 A schematic diagram of the structure of a judgment device for AI-generated academic text provided in an embodiment of the present disclosure is shown;

[0024] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present disclosure clearer, the technical solutions of the present disclosure will be clearly and completely described below in combination with the specific embodiments of the present disclosure and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0026] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such usage is interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "including" and its variants are to be interpreted as open-ended terms meaning "including but not limited to".

[0028] As mentioned above, in the prior art, the common method for detecting whether an academic text is generated by AI mainly focuses on automatic classification by using the text feature differences between academic content generated by AI and academic content written by humans. Specifically, there are the following detection methods: (1) Extracting features such as punctuation, word preference, paragraph distribution, etc. in the text, and applying traditional machine learning methods to classify and analyze the statistical features of academic texts generated by AI and written by humans. Through statistical analysis of human text and AI-generated text, it can be found that human text and AI-generated text have obvious differences in word usage habits, named entity expressions, vocabulary richness, and sentence structure. Therefore, the statistical features of text fragments, such as high-frequency words, stop words, conjunctions, sentence length, paragraph distribution, and other fine-grained text statistical features, are used as feature sets to distinguish human text from AI-generated text. (2) Model verification. When generating text content, the large model calculates the probability distribution of each possible word based on the generated text content. At the same time, in order to ensure the coherence and logic of the generated content, the model usually selects words with a higher sampling probability. Therefore, the autoregressive model with the same architecture is used to calculate whether the probability characteristics of all texts in the text generation process are consistent with the predictions of the AI ​​generation model. (3) Deep learning model classification: directly use AI-generated academic content and human-written content as input to train the deep learning model. (4) Train the language model based solely on a large amount of human-written academic content, and then calculate the differences in the threshold performance of AI-generated academic content and human-written content in the language model to distinguish AI-generated academic content from human-written content.

[0029] However, academic texts in different disciplines are quite different. Therefore, when using existing academic content judgment methods to identify academic texts generated by AI, if you want to achieve a high recognition accuracy in each subject area, you need to train different models or classification methods that are suitable for different disciplines, which will require more resources and energy. The recognition effect of a single model on academic content in different disciplines also varies greatly. Based on this, the present invention proposes a judgment method, device and related products for AI-generated academic texts. The present disclosure is described in detail below through specific embodiments.

[0030] To facilitate understanding of this embodiment, a method for determining whether an AI-generated academic text is disclosed in the embodiment of the present disclosure is first introduced in detail. The execution subject of the method for determining whether an AI-generated academic text is provided in the embodiment of the present disclosure is generally a computer device with certain computing capabilities, and the computer device includes, for example: a terminal device or a server or other processing device, and the terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a personal digital assistant (PDA), etc. In some possible implementations, the method for determining whether an AI-generated academic text is generated may be implemented by a processor calling computer-readable instructions stored in a memory.

[0031] Figure 1 A schematic diagram of the process of judging the AI-generated academic text provided by the embodiment of the present disclosure is shown, from Figure 1 It can be seen that the embodiment of the present disclosure at least includes steps S101-S105:

[0032] S101: Classify the acquired academic texts to be judged to obtain target subject categories.

[0033] This embodiment does not limit the content, length, source, etc. of the academic text to be judged. The academic text to be judged can be any academic-related text, such as all or part of an academic paper, research report, conference paper, case study, etc. The academic text to be judged can come from the Internet world or from user input.

[0034] During implementation, for example, the academic text to be judged can be input by the user through a front-end interface (such as a web page, a mobile application interface, a desktop application interface, etc.), and the user terminal directly or indirectly sends the academic text to be judged input by the user to the execution entity of this embodiment (such as a server) through wired or wireless communication.

[0035] After obtaining the academic text to be judged, the academic text to be judged is classified to obtain one or more target subject categories CID = {CID1, CID2, ..., CID n}, where n is a natural number. In implementation, any academic classification method can be used, such as the Chinese Library Classification Method, Tongfang Knowledge Network 168 Subject Classification, etc. In specific implementation, existing large models, text classification tools, etc. can be used for classification.

[0036] S102: Rewrite the academic text to be judged using a first preset large model and preset rewriting instructions to obtain at least one rewritten text; the preset rewriting instructions include the following rewriting requirements: maintaining the original meaning and retaining the original style.

[0037] In this step, the preset rewriting instructions are limited to include two rewriting requirements: keeping the original meaning and retaining the original style. Among them, the original style includes many unique features of the text, such as word usage habits, sentence structure, tone, etc. Texts written by humans and texts written by AI have their own unique styles. This embodiment uses preset rewriting instructions to enable the model to retain the original style when rewriting the text, avoiding the introduction of too many new changes and uncertainties, so that the subsequent information difference between the academic text to be judged and the rewritten text can accurately reflect the characteristics of the academic text to be judged, and then use the subsequent information difference distribution data to accurately identify the source of the academic text to be judged. The preset rewriting instructions, for example, can be "rewrite text X, keep the original meaning, and retain the original style. Text X:..." Here, the first preset large model can be any existing large language model, which is not limited in this embodiment.

[0038] S103: Calculate the target information volume data of the academic text to be judged and the average information volume data of the at least one rewritten text; subtract the target information volume data from the average information volume data to obtain a target information volume difference.

[0039] Here, information content data is used to measure the uncertainty or randomness of the text, which can reflect the amount of information contained in the text.

[0040] S104: Based on the pre-constructed first difference distribution data, the first AI-generated vote number and the first human-written vote number are calculated using the target subject category and the standardized target information difference; the first difference distribution data reflects the distribution of the information difference after rewriting the AI-generated sample text and the information difference after rewriting the human-written sample text in different subject categories in different intervals.

[0041] Exemplarily, it is assumed that the target subject categories include subject category 1 and subject category 2, and the target information difference after standardization is a. Under subject category 1 in the pre-constructed first difference distribution data, a difference interval A1 including a can be found, and the AI ​​generated sample text rewritten information difference quantity 1 and the human written sample text rewritten information difference quantity 1 in this interval can be determined based on the difference interval A1; under subject category 2 in the first difference distribution data, a difference interval A2 including a can be found, and the AI ​​generated sample text rewritten information difference quantity 2 and the human written sample text rewritten information difference quantity 2 in this interval can be determined based on the difference interval A2. The AI ​​generated sample text rewritten information difference quantities 1 and 2 are added to obtain the first AI generated vote number; the human written vote number is added to obtain the first human written vote number.

[0042] S105: Based on the first AI generated vote number and the first human written vote number, calculate the possibility score of the academic text to be judged being an AI generated text; and based on the possibility score and a preset threshold, judge whether the academic text to be judged is generated by AI to obtain a judgment result.

[0043] In specific implementation, for example, the probability score of whether the academic text to be judged is AI-generated text can be calculated according to the following formula (1):

[0044]

[0045] Among them, SCORE is the possibility score of the academic text to be judged as AI-generated text, CAN is the number of votes generated by the first AI, and CBN is the number of votes written by the first human.

[0046] After obtaining the possibility score, the relationship between the possibility score and the preset threshold can be determined. If the possibility score is greater than or equal to the preset threshold, the judgment result is that the academic text to be judged is generated by AI; if the possibility score is less than the preset threshold, the judgment result is that the academic text to be judged is written by humans. Here, the preset threshold can be set according to actual needs, and this embodiment does not limit this. For example, the preset threshold can be set to 0.85.

[0047] During implementation, the judgment result may be displayed on the front-end page, and may also be displayed on the front-end page together with the possibility score.

[0048] from Figure 1It can be seen from the method shown that the disclosed embodiment uses the pre-constructed first difference distribution data to identify the source of the academic text to be judged. The first difference distribution data is equivalent to a reference standard or feature pattern, which reflects the difference in information before and after rewriting between AI-generated sample texts and human-written sample texts in a large number of different subject categories, and the distribution law in different intervals reflects the characteristic difference in information expression between AI-generated texts and human-written texts. During implementation, according to the target subject category corresponding to the academic text to be judged and the standardized target information difference, the total number of information difference after rewriting of the AI-generated sample text and the total number of information difference after rewriting of the human-written sample text can be obtained under the standard given by the first difference distribution data, and these two total numbers are respectively used as the first AI-generated vote number and the first human-written vote number for the academic text to be judged. Then, the possibility score of the academic text to be judged as an AI-generated text is calculated based on these votes, and then judged in combination with the preset threshold, thereby achieving the purpose of accurately distinguishing whether the academic text is generated by AI. Compared with the prior art, the technical solution provided by the embodiments of the present disclosure does not require separate model training for each subject category. By using a difference interval distribution data, it can accurately determine whether the academic text to be judged is generated by AI, while also greatly saving time and labor costs.

[0049] Further, in order to better illustrate the process of the above-mentioned AI-generated academic text judgment method, as a refinement and extension of the above-mentioned embodiment, the embodiment of the present invention provides several optional embodiments, but is not limited thereto, as shown below:

[0050] In a possible implementation, the method further includes: acquiring academic texts of multiple subject categories written by humans to form a set of academic texts written by humans; for any target subject category, using a second preset macromodel and a target generation instruction corresponding to the target subject category to generate multiple AI-generated academic texts; the AI-generated academic texts corresponding to each target subject category constitute a set of AI-generated academic texts; using the first preset macromodel and the preset rewriting instruction to rewrite any text in the set of academic texts written by humans and the set of academic texts generated by AI to obtain a set of academic texts rewritten by humans and a set of academic texts rewritten by AI; calculating the information volume data of any first target text in the set of academic texts written by humans, as well as the information volume data of the human-rewritten academic text The first information difference value set is obtained by calculating the difference between the average information data of at least one rewritten text in the set corresponding to the first target text; and the information data of any second target text in the AI ​​generated academic text set and the difference between the average information data of at least one rewritten text in the AI ​​rewritten academic text set corresponding to the second target text are calculated to obtain a second information difference value set; the elements in the first information difference value set and the second information difference value set are standardized respectively to obtain a standardized first information difference value set and a standardized second information difference value set; the first difference distribution data is generated according to the standardized first information difference set and the standardized second information difference set.

[0051] In the present embodiment, multiple subject categories include the aforementioned target subject categories. Multiple subject categories can be full subject categories. The subject classification method is the same as the classification method used in the aforementioned S101, and will not be repeated here. During implementation, multiple academic content texts of a certain length can be randomly extracted from the academic content (academic papers, conference papers, dissertations, etc.) under each subject category to form a human-written academic text set. If the subject category total is CN, each subject category has Num texts, then the human-written academic text set includes Num*CN texts.

[0052] During implementation, a comparison table of text serial numbers and subject category IDs (Table 1) can also be generated simultaneously to facilitate subsequent generation of Table 2. The text serial number indicates the position of the text in the collection.

[0053] Table 1 Comparison table of text serial numbers and subject categories

[0054]

[0055]

[0056] Then, for any target subject category, a second preset large model and a target generation instruction corresponding to the target subject category can be used to generate multiple AI-generated academic texts, thereby obtaining an AI-generated academic text set. Here, the second preset large model can be any existing large language model, which is not limited in this embodiment. The first preset large model and the second preset large model can be the same or different. If the total number of subject categories is CN, and there are Num texts under each subject category, then the AI-generated academic text set includes Num*CN texts. During implementation, similarly, a text sequence number and subject category ID comparison table can also be generated for subsequent calculations.

[0057] The target generation instruction may be, for example, "generate a piece of academic text under subject category X". In one possible implementation, the target generation instruction is generated based on subject category, research topic, and academic content type, and the academic content type includes at least one of the following: introduction, research background, research method, and research conclusion.

[0058] For example, the target generation instruction is "generate a section of Z academic content with research topic Y in subject area X. Requirements: (1) content length is MinLen; (2) content complies with academic standards", where X represents the name of a specific subject category; Y represents the research topic under the subject; and Z represents different academic contents such as introduction, research background, research methods, and research conclusions.

[0059] The target generation instructions in this embodiment clearly define the subject categories, research topics, and academic content types, so that the generated AI-generated academic text collection is more in line with the specific needs of actual application scenarios, with greater pertinence, richness, and practicality, thereby significantly improving the accuracy of the first difference distribution data obtained subsequently.

[0060] Then, the first preset large model can be used to generate multiple rewritten texts for each academic text in the set of academic texts written by humans, and a set of academic texts rewritten by humans can be obtained. Let BM represent the set of academic texts rewritten by humans. Suppose there are Num*CN texts in the set of academic texts written by humans, then BM = {{B1M1, B1M2, ..., B1Mr}, {B2M1, B2M2, ..., B2Mr}, ...}, where there are Num*CN subsets in BM, and each subset has r texts. Using the first preset large model, multiple rewritten texts are generated for each academic text in the set of academic texts generated by AI, and a set of academic texts rewritten by AI is obtained. Let AM represent the AI ​​rewritten academic text set. Suppose there are Num*CN texts in the AI ​​rewritten academic text set, then AM = {{A1M1, A1M2, ..., A1Mr}, {A2M1, A2M2, ..., A2Mr}, ...}, where there are Num*CN subsets in AM and each subset has r texts.

[0061] In a possible implementation, the type of information volume data is: information entropy, vocabulary richness or sentence complexity.

[0062] In specific implementation, if the type of information volume data is information entropy, the information volume data of a text can be calculated according to the following formula (2):

[0063]

[0064] Here, we can first count the frequency of occurrence of each word (or character) in the text. Assume that there are n different words (or characters) in the sentence, denoted as W1, W2, ..., Wn, and their frequencies of occurrence are f1, f2, ..., fn. Transform the word frequency into a probability distribution. For each word Wi, its probability p(x i ) can be expressed as formula (3):

[0065]

[0066] The denominator in formula (3) is the total frequency of all words (or characters) in the text.

[0067] If the type of information data is vocabulary richness, the information data of a text can be calculated according to the following method: count the number of different words (characters) appearing in the text, recorded as V; count the total number of words (characters) in the text, recorded as T; V / T is used as the information data of the text. If the type of information data is sentence complexity, the information data of a text can be calculated according to the following method: measure the length of each sentence (which can be the number of words or characters). Assume that there are m sentences in the text, and the length of each sentence is L1, L2, ..., Lm; information data = (L1 + L2 + ... + Lm) / m.

[0068] Exemplarily, if the type of information data is information entropy, the information entropy of any first target text in the set of academic texts written by humans can be calculated using the above formula (2). Let B = {B1, B2, ...} represent the set of academic texts written by humans, then the information entropy set HB = (HB1, HB2, ...) corresponding to the set B can be obtained, where HBi = H(bi). Let the set BM {{B1M1, B1M2 ...}, {B2M1, B2M2 ...} ...} represent the set of academic texts rewritten by humans, where {BiM1, BiM2 ...} corresponds to Bi in the set B. The above formula (2) can be used to calculate the average information entropy of at least one rewritten text corresponding to each first target text, and obtain the average information entropy set HBM = {HB1M, HB2M, ...}, HBiM = Mean (H ({BiM1, BiM2, ...})) = Mean (H (BiM1), H (BiM2), ...). Finally, the corresponding elements of the set HB and the set HBM are subtracted to obtain the first information difference set. If BX is used to represent the first information difference set, then BX = {x1, x2, ...} = HB-HBM = {HB1-HB1M, HB2-HB2M, ...}, and the number of elements in BX is the same as the number of elements in the set of academic texts written by humans. By analogy, the information entropy set HA corresponding to the AI ​​generated academic text set A, the average information entropy set HAM corresponding to the AI ​​rewritten academic text set AM, and the first information difference set AX = {x1, x2, ...} = HA-HAM = {HA1-HA1M, HA2-HA2M, ...}, and the number of elements in AX is the same as the number of elements in the AI ​​generated academic text set.

[0069] In order to ensure that each information difference is within a reasonable range, it is convenient for subsequent data processing and analysis. In this embodiment, the elements in the first information difference set and the second information difference set can be standardized to obtain a standardized first information difference set and a standardized second information difference set. For example, the elements in the set can be standardized using a zero mean standardization method. In a possible implementation of the present disclosure, the elements in the first information difference set and the second information difference set are standardized to obtain a standardized first information difference set and a standardized second information difference set, including: obtaining the maximum difference and the minimum difference in the first information difference set and the second information difference set; for any target difference in the first information difference set and the second information difference set, the target difference is standardized using the maximum difference and the minimum difference to obtain a standardized target difference; each standardized target difference constitutes the standardized first information difference set and the standardized second information difference set.

[0070] In specific implementation, the standardized target difference can be calculated according to the following formula (4):

[0071]

[0072] Among them, x′ represents the standardized target difference, x represents the target difference, minV represents the minimum difference, and MaxV represents the maximum difference.

[0073] During implementation, if x is greater than the maximum difference, then x=maximum difference; if x is less than the minimum difference, then x=minimum difference.

[0074] Finally, according to the two standardized sets, the distribution data of the information difference between the AI-generated text and the human-written text in different intervals after rewriting under different subject categories is generated, that is, the first difference distribution data. In a possible implementation, the first difference distribution data is generated according to the standardized first information difference set and the standardized second information difference set, including: grouping the elements in the standardized first information difference set and the standardized second information difference set according to subject categories to obtain multiple groups of information differences; for any target information difference group, according to the first preset difference interval set, the elements in the target information difference group are divided into intervals to obtain the first total number of differences belonging to the standardized first information difference set and the second total number of differences belonging to the standardized second information difference set under each difference interval; based on each of the subject categories, each of the difference intervals and the first total number and the second total number corresponding to each other, the first difference distribution data is obtained.

[0075] Assume that the standardized first information difference set is BX'={x'1,x'2,x'3,x'4,x'5,x'6,x'7,x'8,x'9,x'10}, and the standardized second information difference set is AX'={y'1,y'2,y'3,y'4,y'5,y'6,y'7,y'8,y'9,y'10}. Among them, the text corresponding to x'1,x'2,y'1,y'2 belongs to subject category 1; the text corresponding to x'3,x'4,y'3,y'4 belongs to subject category 2; the text corresponding to x'5,x'6,y'5,y'6 belongs to subject category 3; the text corresponding to x'7,x'8,y'7,y'8 belongs to subject category 4; the text corresponding to x'9,x'10,y'9,y'10 belongs to subject category 5. The following is an exemplary description of this embodiment in combination with the set BX' and the set AX'.

[0076] First, the elements in sets BX' and AX' are grouped according to subject categories to obtain multiple groups of information differences: information difference group 1 (x'1, x'2, y'1, y'2), information difference group 2 (x'3, x'4, y'3, y'4), information difference group 3 (x'5, x'6, y'5, y'6), information difference group 4 (x'7, x'8, y'7, y'8), and information difference group 5 (x'9, x'10, y'9, y'10).

[0077] Then, for any target information difference value group, the elements in the target information difference value group can be divided into intervals according to the first preset difference interval set, and the total number of differences belonging to the standardized first information difference value set (first total number) and the total number of differences belonging to the standardized second information difference value set (second total number) under each difference interval corresponding to the target information difference value group are obtained. Here, the first preset difference interval set can be set according to actual needs. For example, the first preset difference interval set is set to: [0-0.1], [0.1-0.2], [0.2-0.3], [0.3-0.4], [0.4-0.5], [0.5-0.6], [0.6-0.7], [0.7-0.8], [0.8-0.9], [0.9-1]. For example, suppose that in information difference group 1, x'1=0.85, x'2=0.8, y'1=0.25, y'2=0.24, and according to the first preset difference interval set, the elements in information difference group 1 are divided into intervals, and the first total number and the second total number in each difference interval can be statistically obtained: in the difference interval [0.2-0.3], the first total number is 0, and the second total number is 2; in the difference interval [0.8-0.9], the first total number is 2, and the second total number is 0; in other difference intervals, the first total number and the second total number are both 0. By analogy, the first total number and the second total number in each difference interval corresponding to information difference group 2, information difference group 3, information difference group 4, and information difference group 5 can be obtained.

[0078] Finally, based on each subject category, difference interval and the first total number and the second total number respectively corresponding to each other, the first difference distribution data is obtained. For example, the following Table 2 is the first difference distribution data.

[0079] Table 2 First difference distribution data

[0080]

[0081]

[0082] This embodiment rewrites a large number of known human-written and AI-generated texts, and calculates the difference between the amount of text information before and after the rewriting to obtain difference distribution data, which can reflect the general rules and characteristics of these two types of texts after specific processing. The difference distribution data can be used to accurately determine the source of new academic texts. The information difference calculation is small, the distribution data is fast to build, the efficiency is high, and the construction cost is low.

[0083] In actual application scenarios, if the academic text to be judged has problems such as not having obvious disciplinary tendencies, spanning multiple disciplinary fields, or being too new to accurately judge the disciplinary category, then the accuracy of the probability score calculated using the above formula (1) is poor. Based on this, in a possible implementation of the present disclosure, the method further includes:

[0084] According to the second preset difference interval set, the elements in the standardized first information amount difference set and the standardized second information amount difference set are divided into intervals to obtain a third total number of differences belonging to the standardized first information amount difference set and a fourth total number of differences belonging to the standardized second information amount difference set in each interval;

[0085] Based on the second preset difference interval set, the third total number and the fourth total number corresponding to each interval, second difference distribution data is obtained, where the second difference distribution data reflects the distribution of the text information amount difference between the AI ​​generated text and the human written text after rewriting in different intervals;

[0086] Based on the second difference distribution data, the standardized target information amount difference is used to calculate the second AI generated vote number and the second human written vote number;

[0087] The method of calculating the possibility score of the academic text to be judged to be an AI-generated text based on the first AI-generated vote number and the first human-written vote number includes: calculating a first possibility score according to the first AI-generated vote number, the first human-written vote number and a preset first weight; calculating a second possibility score according to the second AI-generated vote number, the second human-written vote number and a preset second weight; and adding the first possibility score and the second possibility score to obtain the possibility score of the academic text to be judged to be AI-generated.

[0088] In this embodiment, for any target difference interval in the second preset difference interval set, the total number of differences belonging to the target difference interval in the standardized first information amount difference set can be counted to obtain a third total number corresponding to the target difference interval, and the total number of differences belonging to the target difference interval in the standardized second information amount difference set can be counted to obtain a fourth total number corresponding to the target difference interval. By analogy, the third total number and the fourth total number corresponding to each difference interval can be obtained.

[0089] Then, based on the second preset difference interval set, the third total number and the fourth total number corresponding to each interval, the second difference distribution data is obtained. Exemplarily, the following Table 3 is the second difference distribution data.

[0090] Table 3 Second difference distribution data

[0091] Difference interval Fourth total The third total 0-0.1 50086 267 0.1-0.2 408960 5341 ... ... ... 0.9-1.0 178 50889

[0092] Next, based on the second difference distribution data, the second AI generated vote number and the second human written vote number are calculated using the standardized target information amount difference. Exemplarily, the standardized target information amount difference is set to b. The difference interval B including b can be determined in the second difference distribution data, and the number of information amount differences after the AI ​​generated sample text is rewritten (the second AI generated vote number) and the number of information amount differences after the human written sample text is rewritten (the second human written vote number) can be determined according to the difference interval B.

[0093] Finally, the first possibility score is calculated based on the first AI-generated vote count, the first human-written vote count, and the preset first weight; the second possibility score is calculated based on the second AI-generated vote count, the second human-written vote count, and the preset second weight; the first possibility score and the second possibility score are added together to obtain the possibility score of the academic text to be judged to be generated by AI. That is, the possibility score of the academic text to be judged to be generated by AI is calculated according to the following formula (5):

[0094]

[0095] Wherein, W is a preset first weight, 1-W is a preset second weight, AN is a fourth total, and BNw is a third total. In implementation, W is greater than 0.5 and less than or equal to 1, and can be set according to actual needs, and is not limited in this embodiment.

[0096] This embodiment constructs the second difference distribution data, which is based on the first difference distribution data and ignores the subject category. During implementation, based on the second difference distribution data, the second AI-generated votes and the second human-written votes can be calculated using the standardized target information difference; when calculating the probability score of the academic text to be judged to be AI-generated, the second AI-generated votes and the second human-written votes are also taken into account, so that when the academic text to be judged does not have an obvious subject tendency, spans multiple subject areas, or the subject category is difficult to accurately judge, the accuracy of the final probability score can be guaranteed.

[0097] The present disclosure also provides a method for determining whether an AI-generated academic text is generated, comprising the following steps:

[0098] Step S1: Acquire academic texts of various subject categories written by humans to form a collection of academic texts written by humans.

[0099] Step S2: For any target subject category, use the second preset large model and the target generation instructions corresponding to the target subject category to generate an AI-generated academic text set.

[0100] During implementation, goal generation instructions are generated based on subject categories, research topics, and academic content types, which include introduction, research background, research methods, and research conclusions.

[0101] Step S3: Rewrite any text in the human-written academic text set and the AI-generated academic text set by using the first preset large model and the preset rewriting instructions to obtain the human-rewritten academic text set and the AI-rewritten academic text set.

[0102] Step S4: Calculate the difference between the information entropy of any first target text in the set of academic texts written by humans and the average information entropy of at least one rewritten text in the set of academic texts rewritten by humans and corresponding to the first target text, and obtain a first information difference set.

[0103] Step S5: Calculate the information entropy of any second target text in the AI ​​generated academic text set, and the difference between the average information entropy of at least one rewritten text in the AI ​​rewritten academic text set corresponding to the second target text, to obtain a second information difference set.

[0104] Step S6: Obtain the maximum difference and the minimum difference in the first information amount difference set and the second information amount difference set.

[0105] Step S7: For any target difference in the first information difference set and the second information difference set, the target difference is standardized using the maximum difference and the minimum difference to obtain a standardized target difference; each standardized target difference constitutes a standardized first information difference set and a standardized second information difference set.

[0106] Step S8: Group the elements in the standardized first information difference value set and the standardized second information difference value set according to subject categories to obtain multiple groups of information difference values.

[0107] Step S9: For any target information difference group, the elements in the target information difference group are divided into intervals according to the first preset difference interval set, and a first total number of differences belonging to the standardized first information difference set and a second total number of differences belonging to the standardized second information difference set are obtained under each difference interval.

[0108] Step S10: Based on each subject category, each difference interval and the corresponding first total number and second total number, obtain first difference distribution data.

[0109] Step S11: According to the second preset difference interval set, the elements in the standardized first information difference set and the standardized second information difference set are divided into intervals to obtain the third total number of differences belonging to the standardized first information difference set and the fourth total number of differences belonging to the standardized second information difference set in each interval.

[0110] Step S12: Based on the second preset difference interval set, the third total number and the fourth total number corresponding to each interval, obtain second difference distribution data.

[0111] Step S13: Classify the acquired academic text to be judged to obtain the target subject category.

[0112] Step S14: rewrite the academic text to be judged using the first preset large model and preset rewriting instructions to obtain at least one rewritten text.

[0113] Step S15: Calculate the target information volume data of the academic text to be judged and the average information volume data of at least one rewritten text; subtract the target information volume data from the average information volume data to obtain the target information volume difference.

[0114] Step S16: Based on the pre-constructed first difference distribution data, using the target subject category and the standardized target information difference, calculate the first AI-generated vote number and the first human-written vote number.

[0115] Step S17: Based on the second difference distribution data, using the standardized target information difference, calculate the second AI generated vote number and the second human written vote number.

[0116] Step S18: Calculate a first possibility score based on the number of votes generated by the first AI, the number of votes written by the first human, and the preset first weight.

[0117] Step S19: Calculate a second possibility score based on the number of votes generated by the second AI, the number of votes written by the second human, and the preset second weight.

[0118] Step S20: Add the first possibility score and the second possibility score to obtain a possibility score that the academic text to be determined is generated by AI.

[0119] Step S21: Based on the possibility score and the preset threshold, determine whether the academic text to be judged is generated by AI and obtain a judgment result.

[0120] Those skilled in the art will appreciate that, in the above method of the specific implementation manner, the writing order of the steps does not mean a strict execution order, but does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0121] It should be noted that, in practical applications, all possible implementations described above can be combined in any manner to form possible embodiments of the present disclosure, which will not be described one by one here.

[0122] Based on the same concept, the present disclosure also provides a judgment device for AI-generated academic texts. Figure 2 A schematic diagram of the structure of the judgment device for AI-generated academic text provided by the embodiment of the present disclosure is shown, see Figure 2 As shown, the judgment device 200 for AI-generated academic text provided in the embodiment of the present disclosure includes:

[0123] The classification module 201 is used to classify the acquired academic text to be judged to obtain the target subject category;

[0124] The rewriting module 202 is used to rewrite the academic text to be judged by using the first preset large model and the preset rewriting instruction to obtain at least one rewritten text; the preset rewriting instruction includes the following rewriting requirements: maintaining the original meaning and retaining the original style;

[0125] The calculation module 203 is used to calculate the target information amount data of the academic text to be judged and the average information amount data of the at least one rewritten text; subtract the target information amount data from the average information amount data to obtain a target information amount difference;

[0126] The voting module 204 is used to calculate the first AI-generated vote number and the first human-written vote number based on the pre-constructed first difference distribution data, using the target subject category and the standardized target information difference; the first difference distribution data reflects the distribution of the information difference of the AI-generated sample text after rewriting and the information difference of the human-written sample text after rewriting in different subject categories in different intervals;

[0127] The judgment module 205 is used to calculate the possibility score of the academic text to be judged being an AI-generated text based on the first AI-generated vote number and the first human-written vote number; and judge whether the academic text to be judged is generated by AI based on the possibility score and a preset threshold to obtain a judgment result.

[0128] In a possible implementation, the device further includes a first construction module, which is used to: obtain academic texts of multiple subject categories written by humans to form a set of academic texts written by humans; for any target subject category, generate multiple AI-generated academic texts using a second preset macromodel and a target generation instruction corresponding to the target subject category; the AI-generated academic texts corresponding to each target subject category constitute a set of AI-generated academic texts; rewrite any text in the set of academic texts written by humans and the set of academic texts generated by AI using the first preset macromodel and the preset rewriting instruction to obtain a set of academic texts rewritten by humans and a set of academic texts rewritten by AI; calculate the information volume data of any first target text in the set of academic texts written by humans, and the information volume data of the human-rewritten academic texts. The first information difference set is obtained by calculating the difference between the average information data of at least one rewritten text corresponding to the first target text in the academic text set generated by the AI, and the difference between the average information data of at least one rewritten text corresponding to the second target text in the academic text set rewritten by the AI, to obtain the second information difference set; the elements in the first information difference set and the second information difference set are standardized respectively to obtain a standardized first information difference set and a standardized second information difference set; the first difference distribution data is generated according to the standardized first information difference set and the standardized second information difference set.

[0129] In a possible implementation, in the above-mentioned device, the target generation instruction is generated based on subject category, research topic, and academic content type, and the academic content type includes at least one of the following: introduction, research background, research method, and research conclusion.

[0130] In a possible implementation, in the above device, the type of information volume data is: information entropy, vocabulary richness or sentence complexity.

[0131] In a possible implementation, the device also includes a standardization processing module, which is used to: obtain the maximum difference and the minimum difference in the first information amount difference value set and the second information amount difference value set; for any target difference in the first information amount difference value set and the second information amount difference value set, use the maximum difference and the minimum difference to standardize the target difference to obtain a standardized target difference; each standardized target difference constitutes the standardized first information amount difference value set and the standardized second information amount difference value set.

[0132] In a possible implementation, in the above-mentioned device, the first construction module is used to: group the elements in the standardized first information difference set and the standardized second information difference set according to subject categories to obtain multiple groups of information differences; for any target information difference group, divide the elements in the target information difference group into intervals according to a first preset difference interval set to obtain a first total number of differences belonging to the standardized first information difference set and a second total number of differences belonging to the standardized second information difference set under each difference interval; obtain the first difference distribution data based on each of the subject categories, each of the difference intervals and the respectively corresponding first total number and second total number.

[0133] In a possible implementation, the device further includes a second construction module, which is used to: divide the elements in the standardized first information difference set and the standardized second information difference set into intervals according to a second preset difference interval set, and obtain a third total number of differences belonging to the standardized first information difference set and a fourth total number of differences belonging to the standardized second information difference set in each interval; obtain second difference distribution data based on the second preset difference interval set, the third total number and the fourth total number corresponding to each interval, and the second difference distribution data reflects the distribution of text information difference between AI-generated text and human-written text after rewriting in different intervals; based on the second difference distribution data, use the standardized target information difference to calculate the second AI-generated vote number and the second human-written vote number;

[0134] The judgment module is used to: calculate a first possibility score based on the first AI generated vote number, the first human written vote number and a preset first weight; calculate a second possibility score based on the second AI generated vote number, the second human written vote number and a preset second weight; add the first possibility score and the second possibility score to obtain a possibility score for the academic text to be judged to be AI generated.

[0135] It should be noted that any of the above-mentioned AI-generated academic text judgment devices can implement the above-mentioned AI-generated academic text judgment method one-to-one, which will not be repeated here.

[0136] Figure 3 FIG. 1 is a schematic diagram showing the structure of an electronic device provided by an embodiment of the present disclosure. Figure 3As shown, at the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. Among them, the memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage, etc. Of course, the electronic device may also include hardware required for other services.

[0137] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0138] The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0139] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a judgment device for AI-generated academic texts at the logical level. The processor executes the program stored in the memory and is specifically used to execute the above method.

[0140] The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present disclosure may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present disclosure may be directly embodied as being executed by a hardware decoding processor, or may be executed by a combination of hardware and software modules in a decoding processor. The software module may be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0141] The electronic device can execute the method for judging whether an AI-generated academic text is provided in the embodiments of the present disclosure, and realize a judgment device for judging whether an AI-generated academic text is generated. Figure 2 The functions of the illustrated embodiments will not be described in detail in the embodiments of the present disclosure.

[0142] The embodiments of the present disclosure also propose a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple application programs, the electronic device can execute the judgment method of AI-generated academic text provided by multiple embodiments of the present disclosure.

[0143] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure 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.

[0144] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0145] 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.

[0146] 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.

[0147] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0148] 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.

[0149] 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, 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.

[0150] The embodiments of the present disclosure also provide a computer program product, which carries a program code. The instructions included in the program code can be used to execute the steps of the method for judging the AI-generated academic text described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0151] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0152] 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 further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity or device including the element.

[0153] It will be appreciated by those skilled in the art that the embodiments of the present disclosure may be provided as methods, systems or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present disclosure 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.

[0154] The above are only embodiments of the present disclosure and are not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the scope of the claims of the present disclosure.

Claims

1. A method for judging AI-generated academic texts, characterized in that: The method comprises: Classify the acquired academic texts to be judged to obtain the target subject category; Rewrite the academic text to be judged by using the first preset large model and the preset rewriting instruction to obtain at least one rewritten text; the preset rewriting instruction includes the following rewriting requirements: maintaining the original meaning and retaining the original style; Calculating the target information amount data of the academic text to be judged and the average information amount data of the at least one rewritten text; subtracting the target information amount data from the average information amount data to obtain a target information amount difference; Based on the pre-constructed first difference distribution data, the first AI-generated vote number and the first human-written vote number are calculated using the target subject category and the standardized target information difference; the first difference distribution data reflects the distribution of the information difference of the AI-generated sample text after rewriting and the information difference of the human-written sample text after rewriting in different subject categories in different intervals; Based on the first AI generated vote number and the first human written vote number, a possibility score of the academic text to be judged being an AI generated text is calculated; and based on the possibility score and a preset threshold, it is judged whether the academic text to be judged is generated by AI to obtain a judgment result.

2. The method according to claim 1, characterized in that The method further comprises: Acquire academic texts of various subject categories written by humans to form a collection of academic texts written by humans; For any target subject category, a plurality of AI-generated academic texts are generated using the second preset macromodel and the target generation instructions corresponding to the target subject category; the AI-generated academic texts corresponding to each target subject category constitute a set of AI-generated academic texts; Rewrite any text in the human-written academic text set and the AI-generated academic text set by using the first preset macromodel and the preset rewriting instruction to obtain a human-rewritten academic text set and an AI-rewritten academic text set; Calculating the difference between the information content data of any first target text in the set of human-written academic texts and the average information content data of at least one rewritten text in the set of human-rewritten academic texts corresponding to the first target text, to obtain a first information content difference set; and Calculate the difference between the information content data of any second target text in the AI-generated academic text set and the average information content data of at least one rewritten text in the AI-rewritten academic text set corresponding to the second target text, to obtain a second information content difference set; Standardizing the elements in the first information amount difference value set and the second information amount difference value set respectively to obtain a standardized first information amount difference value set and a standardized second information amount difference value set; The first difference distribution data is generated according to the standardized first information amount difference value set and the standardized second information amount difference value set.

3. The method according to claim 2, characterized in that The target generation instruction is generated based on subject category, research topic, and academic content type, and the academic content type includes at least one of the following: introduction, research background, research method, and research conclusion.

4. The method according to claim 1, characterized in that: The types of information data are: information entropy, vocabulary richness or sentence complexity.

5. The method according to claim 2, characterized in that: The step of respectively standardizing the elements in the first information amount difference set and the second information amount difference set to obtain a standardized first information amount difference set and a standardized second information amount difference set includes: Obtaining the maximum difference and the minimum difference in the first information amount difference set and the second information amount difference set; For any target difference in the first information amount difference set and the second information amount difference set, the target difference is standardized using the maximum difference and the minimum difference to obtain a standardized target difference; each standardized target difference constitutes the standardized first information amount difference set and the standardized second information amount difference set.

6. The method according to claim 2, characterized in that The step of generating the first difference distribution data according to the standardized first information amount difference value set and the standardized second information amount difference value set includes: According to subject categories, the elements in the standardized first information difference value set and the standardized second information difference value set are grouped to obtain multiple groups of information difference values; For any target information amount difference value group, according to the first preset difference value interval set, the elements in the target information amount difference value group are divided into intervals, and a first total number of differences belonging to the standardized first information amount difference value set and a second total number of differences belonging to the standardized second information amount difference value set are obtained under each difference interval; Based on each of the subject categories, each of the difference intervals and the respectively corresponding first totals and second totals, the first difference distribution data is obtained.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: According to the second preset difference interval set, the elements in the standardized first information amount difference set and the standardized second information amount difference set are divided into intervals to obtain a third total number of differences belonging to the standardized first information amount difference set and a fourth total number of differences belonging to the standardized second information amount difference set in each interval; Based on the second preset difference interval set, the third total number and the fourth total number corresponding to each interval, second difference distribution data is obtained, where the second difference distribution data reflects the distribution of the text information amount difference between the AI ​​generated text and the human written text after rewriting in different intervals; Based on the second difference distribution data, the standardized target information amount difference is used to calculate the second AI generated vote number and the second human written vote number; The calculating, based on the first AI generated vote number and the first human written vote number, a probability score of the to-be-determined academic text being an AI generated text comprises: Calculate a first possibility score according to the number of votes generated by the first AI, the number of votes written by the first human, and a preset first weight; Calculate a second possibility score according to the number of votes generated by the second AI, the number of votes written by the second human, and a preset second weight; The first possibility score and the second possibility score are added together to obtain a possibility score of whether the academic text to be determined is generated by AI.

8. A judgment device for AI-generated academic texts, characterized in that: The device comprises: The classification module is used to classify the acquired academic texts to be judged and obtain the target subject category; A rewriting module is used to rewrite the academic text to be judged by using a first preset large model and a preset rewriting instruction to obtain at least one rewritten text; the preset rewriting instruction includes the following rewriting requirements: maintaining the original meaning and retaining the original style; A calculation module, used for calculating the target information amount data of the academic text to be judged and the average information amount data of the at least one rewritten text; subtracting the target information amount data from the average information amount data to obtain a target information amount difference; A voting module is used to calculate the first AI-generated vote count and the first human-written vote count based on the pre-constructed first difference distribution data, using the target subject category and the standardized target information difference; the first difference distribution data reflects the distribution of the information difference of the AI-generated sample text after rewriting and the information difference of the human-written sample text after rewriting in different subject categories in different intervals; The judgment module is used to calculate the possibility score of the academic text to be judged being an AI-generated text based on the first AI-generated vote number and the first human-written vote number; and judge whether the academic text to be judged is generated by AI based on the possibility score and a preset threshold to obtain a judgment result.

9. An electronic device, comprising: processor; as well as A memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to perform the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing one or more programs, characterized in that: When the one or more programs are executed by an electronic device including a plurality of application programs, the electronic device executes the steps of the method as claimed in any one of claims 1 to 7.

Images