A generative model guided word test method and system

Abstract

The application provides a generative model guided word test method and system. The method comprises the following steps: configuring model parameters of a generative model; using the generative model to translate original prompt words input by a user, obtaining a translation result, and performing synonym or near-synonym processing on the translation result to obtain processed multiple versions of text data information; monitoring, in real time, a use condition of the multiple versions of text data information by the user, evaluating each synonym or near-synonym according to the use condition, and determining a priority corresponding to the synonym or near-synonym through an evaluation result. The system comprises modules corresponding to the method steps.

Description

Technical Field

[0001] This invention proposes a generative model guide word testing method and system, belonging to the field of guide word testing technology. Background Technology

[0002] Currently, for some image generation models (such as Stable Diffusion) and large language models, the prompts need to be refined during fine-tuning tests or when performing certain generation tasks to achieve better results. The prompt refinement process is highly uncertain and difficult to control, thus it is time-consuming. Users often need a lot of time to debug the prompts when performing generation tasks with generative models. Summary of the Invention

[0003] This invention provides a generative model-based word testing method and system to solve the aforementioned technical problems in the prior art. The technical solution adopted is as follows:

[0004] A generative model guide word testing method, the generative model guide word testing method comprising:

[0005] Configure the model parameters of the generative model, wherein the generative model adopts a language-based large model, such as the GPT series model;

[0006] The generative model is used to translate the original prompts input by the user to obtain translation results, and the translation results are processed by synonyms or near-synonyms to obtain multiple versions of text data information after processing.

[0007] The system monitors in real time the usage of synonyms or near-synonyms in the multiple versions of text data, evaluates each synonym or near-synonym based on the usage, and determines the priority of the synonym or near-synonym based on the evaluation results.

[0008] Furthermore, the generative model is used to translate the original prompts input by the user, obtaining translation results. The translation results are then processed using synonyms or near-synonyms to obtain multiple versions of processed text data, including:

[0009] Initialize the generative model with the parameters configured.

[0010] The system receives raw prompts from the user in real time and inputs them into the generative model with the parameters configured.

[0011] The generative model with the completed parameter configuration translates the original prompt words according to the user's target language requirements to obtain the translation result;

[0012] Extract the text data information corresponding to the translation result;

[0013] The words in the text data information corresponding to the translation result are identified to obtain the words contained in the text data information;

[0014] The words contained in the text data are processed to generate synonyms or near-synonyms corresponding to the words.

[0015] The synonyms or near-synonyms are replaced sequentially with their corresponding words to generate multiple versions of text data information with synonyms or near-synonyms that correspond to the text data information.

[0016] Furthermore, the text data information includes synonym or near-synonym processing to generate near-synonyms or near-synonyms corresponding to the words, including:

[0017] Perform semantic analysis on the words contained in the text data to obtain the word meaning and context corresponding to the words;

[0018] Based on the meaning and context of the word, synonyms or near-synonyms are obtained using a synonym or near-synonym algorithm.

[0019] Generate a list of synonyms or near-synonyms corresponding to the given word.

[0020] Furthermore, based on the meaning and context of the word, synonyms or near-synonyms are obtained using a synonym or near-synonym algorithm, including:

[0021] The words are represented by word vectors using a pre-trained language model;

[0022] Word vectors are used to calculate the similarity between a word and other words in a thesaurus / near-synonyms; wherein, the similarity is obtained by the following formula:

[0023]

[0024] Where S represents the similarity score between the word and other words in the thesaurus / near-synonyms; A 01 A represents the word vector corresponding to a word; 02 represents the word vector corresponding to other words in the thesaurus / near-synonyms; 'a' represents the similarity compensation coefficient.

[0025]

[0026] Where n represents the dimension of the word vector corresponding to the word and the word vectors corresponding to other words in the synonym / near-synonym dictionary; x i This represents the i-th element of the word vector corresponding to the word; y i This represents the i-th element of the word vector corresponding to other words in the thesaurus / near-synonyms; c 0i This represents the preset reference data for the i-th vector element;

[0027] Words with a similarity greater than a preset similarity threshold are selected as candidate synonyms / near-synonyms for the given word;

[0028] The candidate synonyms / near-synonyms are converted to generate synonyms / near-synonyms corresponding to the words.

[0029] Furthermore, the system monitors in real-time the usage of synonyms or near-synonyms in the multiple versions of text data, evaluates each synonym or near-synonym based on the usage, and determines the priority of each synonym or near-synonym based on the evaluation results, including:

[0030] The system monitors in real time the user's usage parameters for synonyms or near-synonyms in the text data; wherein, the usage parameters include the number of times and frequency of use of synonyms or near-synonyms.

[0031] The evaluation parameters of the synonyms or near-synonyms are obtained based on their usage parameters; wherein, the evaluation parameters are obtained using the following formula:

[0032]

[0033] Where f represents the evaluation parameter corresponding to the synonym or near-synonym; δ represents the parameter baseline value; N represents the total number of times a synonym or near-synonym has been used; M represents the total number of times all words with the same meaning as a synonym or near-synonym have been used; n represents the number of unit fields entered by the user, and the number of characters contained in the unit field ranges from 300 to 500; ξ represents the first compensation coefficient; σ represents the second compensation coefficient;

[0034] The expression for the reference value δ is as follows:

[0035]

[0036] Among them, C i This represents the number of synonyms or near-synonyms contained in the i-th unit field; C ni This represents the number of times all words in the i-th unit field that have the same meaning as a synonym or near-synonym are used;

[0037] Furthermore, the expression for the first compensation coefficient ξ is as follows:

[0038]

[0039] Meanwhile, the expression for the second compensation coefficient σ is as follows:

[0040]

[0041] Among them, C i-1 This indicates the number of synonyms or near-synonyms contained in the (i-1)th unit field;

[0042] The priorities of the synonyms or near-synonyms are set according to the evaluation parameters from high to low, and the translation result is generated according to the synonym or near-synonym with the highest priority when the user performs the next translation.

[0043] A generative model guide word testing system, the generative model guide word testing system comprising:

[0044] The parameter configuration module is used to configure the model parameters of the generative model, wherein the generative model adopts a model based on a large language model, such as the GPT series model.

[0045] The translation execution module is used to translate the original prompt words input by the user using the generative model, obtain translation results, and process the translation results for synonyms or near-synonyms to obtain multiple versions of text data information after processing.

[0046] The real-time monitoring module is used to monitor the user's usage of synonyms or near-synonyms in the multiple versions of text data information in real time, evaluate each synonym or near-synonym based on the usage, and determine the priority of the synonym or near-synonym based on the evaluation results.

[0047] Furthermore, the translation execution module includes:

[0048] An initialization module is used to initialize the generative model after the parameter configuration is completed;

[0049] The data input module is used to receive the original prompt words input by the user in real time and input the received original prompt words into the generative model with completed parameter configuration;

[0050] The translation control module is used to translate the original prompt words according to the user's target language requirements through the generative model with completed parameter configuration, and obtain the translation result;

[0051] The text data information extraction module is used to extract the text data information corresponding to the translation result;

[0052] The word acquisition module is used to identify words in the text data information corresponding to the translation result and acquire the words contained in the text data information;

[0053] The synonym / near-synonym processing module is used to process the words contained in the text data information into near-synonyms or near-synonyms, and generate near-synonyms or near-synonyms corresponding to the words;

[0054] The replacement operation module is used to sequentially replace the corresponding words with the synonyms or near-synonyms, generating multiple versions of text data information with synonyms or near-synonyms corresponding to the text data information.

[0055] Furthermore, the synonym / near-synonym processing module includes:

[0056] The word meaning analysis module is used to perform word meaning analysis on the words contained in the text data information to obtain the word meaning and context corresponding to the words;

[0057] The synonym or near-synonym acquisition module is used to acquire the synonyms or near-synonyms corresponding to the words based on the meaning and context of the words;

[0058] The list generation module is used to generate a list of synonyms or near-synonyms corresponding to the word.

[0059] Furthermore, the synonym or near-synonym acquisition module includes:

[0060] The word vector representation module is used to represent the words using a pre-trained language model;

[0061] The similarity acquisition module is used to calculate the similarity between a word and other words in the thesaurus / near-synonyms using word vector representation; wherein, the similarity is obtained by the following formula:

[0062]

[0063] Where S represents the similarity score between the word and other words in the thesaurus / near-synonyms; A 01 A represents the word vector corresponding to a word; 02 represents the word vector corresponding to other words in the thesaurus / near-synonyms; 'a' represents the similarity compensation coefficient.

[0064]

[0065] Where n represents the dimension of the word vector corresponding to the word and the word vectors corresponding to other words in the synonym / near-synonym dictionary; xi This represents the i-th element of the word vector corresponding to the word; y i This represents the i-th element of the word vector corresponding to other words in the thesaurus / near-synonyms; c 0i This represents the preset reference data for the i-th vector element;

[0066] The candidate synonym / near-synonym acquisition module is used to select words with a similarity greater than a preset similarity threshold as candidate synonyms / near-synonyms corresponding to the word.

[0067] The synonym / near-synonym generation module is used to convert the candidate synonyms / near-synonyms to generate synonyms / near-synonyms corresponding to the words.

[0068] Furthermore, the real-time monitoring module includes:

[0069] A real-time parameter monitoring module is used to monitor the user's usage parameters for synonyms or near-synonyms in the text data information in real time; wherein, the usage parameters include the number of times and frequency of use of synonyms or near-synonyms;

[0070] The evaluation parameter acquisition module is used to acquire the evaluation parameters of the synonyms or near-synonyms based on their usage parameters; wherein the evaluation parameters are acquired using the following formula:

[0071]

[0072] Where f represents the evaluation parameter corresponding to the synonym or near-synonym; δ represents the parameter baseline value; N represents the total number of times a synonym or near-synonym has been used; M represents the total number of times all words with the same meaning as a synonym or near-synonym have been used; n represents the number of unit fields entered by the user, and the number of characters contained in the unit field ranges from 300 to 500; ξ represents the first compensation coefficient; σ represents the second compensation coefficient;

[0073] The expression for the reference value δ is as follows:

[0074]

[0075] Among them, C i This represents the number of synonyms or near-synonyms contained in the i-th unit field; C ni This represents the number of times all words in the i-th unit field that have the same meaning as a synonym or near-synonym are used;

[0076] Furthermore, the expression for the first compensation coefficient ξ is as follows:

[0077]

[0078] Meanwhile, the expression for the second compensation coefficient σ is as follows:

[0079]

[0080] Among them, C i-1 This indicates the number of synonyms or near-synonyms contained in the (i-1)th unit field;

[0081] The priority processing module is used to set the priority of the synonyms or near-synonyms in descending order of the evaluation parameters, and generate the translation result according to the synonym or near-synonym with the highest priority when the user performs the next translation.

[0082] Beneficial effects of this invention:

[0083] This invention proposes a generative model guidance word testing method and system based on a large language model. It generates adjustable number and language prompts based on user input, translating the user's original prompts into multiple languages ​​for testing the model's stability across different languages. Furthermore, it generates various combinations of synonyms and near-synonyms based on the user's prompts, facilitating rapid refinement of prompt words. Users can easily set the number and language of generated prompts and directly input them into the generative model to obtain results in batches. The prompt refinement process is automated to a certain extent, providing automatic input to the generative model and batch output of generated results, saving users time and improving the generation effect. By generating multiple versions of text data and providing synonym or near-synonym selection, the system can better meet users' personalized needs and enhance user experience. By monitoring and evaluating user behavior in real time, the system can understand users' preferences for synonyms or near-synonyms, thus recommending appropriate words more accurately and improving the accuracy and relevance of the generated text. By collecting and analyzing user feedback and behavioral data, the parameters and algorithms of generative models can be continuously optimized, improving the model's performance and adaptability. Through continuous learning and adaptation to user habits, the system can gradually enhance its intelligence, providing users with more intelligent and personalized services. Attached Figure Description

[0084] Figure 1 This is a flowchart corresponding to the method described in this invention;

[0085] Figure 2 This is a system block diagram corresponding to the system described in this invention. Detailed Implementation

[0086] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0087] This invention proposes a generative model-guided word testing method, such as... Figure 1 As shown, the generative model guide word testing method includes:

[0088] S1. Configure the model parameters of the generative model, wherein the generative model adopts a model based on a large language model, such as the GPT series model.

[0089] S2. The generative model is used to translate the original prompts input by the user to obtain translation results, and the translation results are processed by synonyms or near-synonyms to obtain multiple versions of text data information after processing.

[0090] S3. Monitor in real time the usage of synonyms or near-synonyms in the multiple versions of text data information, evaluate each synonym or near-synonym based on the usage, and determine the priority of the synonym or near-synonym based on the evaluation results.

[0091] The working principle of the above technical solution is as follows: Configure the generative model (S1):

[0092] First, configure the model parameters for the generative model. The generative model here is based on large language models, such as the GPT series. These models are trained on a large amount of text data and are able to generate fluent and context-sensitive text.

[0093] Translate prompt words and process synonyms or near-synonyms (S2):

[0094] Next, the configured generative model is used to translate the original prompts input by the user. The purpose of translation is to transform the original prompts into a form that is easier for the model to understand or generate.

[0095] After translation, synonyms or near-synonyms in the translation results are processed. This processing may include the identification, replacement, or expansion of synonyms or near-synonyms to generate multiple versions of the text data.

[0096] Real-time monitoring and evaluation of usage (S3):

[0097] Real-time monitoring of user usage of synonyms or near-synonyms across multiple versions of text data. This can be achieved by tracking user interactions with the system and analyzing user behavior logs.

[0098] Each synonym or near-synonym is evaluated based on user behavior. The evaluation may be based on multiple dimensions, such as user selection frequency, usage duration, and feedback.

[0099] Based on the evaluation results, the priorities of synonyms or near-synonyms are determined. Synonyms or near-synonyms with higher priorities are more likely to be recommended to users by the system, thereby improving user satisfaction and system accuracy.

[0100] The effects of the above technical solution are as follows: By generating multiple versions of text data and providing the option to choose synonyms or near-synonyms, the system can better meet users' personalized needs and improve user experience. By monitoring and evaluating user behavior in real time, the system can understand users' preferences for synonyms or near-synonyms, thereby recommending appropriate words more accurately and improving the accuracy and relevance of the generated text. By collecting and analyzing user feedback and behavioral data, the parameters and algorithms of the generative model can be continuously optimized, improving the model's performance and adaptability. By continuously learning and adapting to users' usage habits, the system can gradually improve its intelligence, providing users with more intelligent and personalized services.

[0101] In summary, this technical solution, by configuring generative models, translating prompts, handling synonyms or near-synonyms, and monitoring and evaluating user behavior in real time, can improve user experience, increase accuracy, optimize model performance, and enhance system intelligence.

[0102] In one embodiment of the present invention, the generative model is used to translate the original prompt words input by the user to obtain translation results, and the translation results are processed using synonyms or near-synonyms to obtain multiple versions of processed text data information, including:

[0103] S201. Initialize the generative model with completed parameter configuration;

[0104] S202. Receive the original prompt words input by the user in real time, and input the received original prompt words into the generative model with completed parameter configuration;

[0105] S203. The original prompt words are translated according to the user's target language requirements using the generative model with the completed parameter configuration to obtain the translation result;

[0106] S204. Extract the text data information corresponding to the translation result;

[0107] S205. Identify the words in the text data information corresponding to the translation result, and obtain the words contained in the text data information;

[0108] S206. Process the words contained in the text data information for synonyms or near-synonyms to generate near-synonyms or near-synonyms corresponding to the words;

[0109] S207. Replace the corresponding words with the synonyms or near-synonyms in sequence to generate multiple versions of text data information with synonyms or near-synonyms corresponding to the text data information.

[0110] The working principle of the above technical solution is as follows: Initialize the generative model (S201):

[0111] Before starting the translation, the generative model with the parameters configured is first initialized to ensure that the model is in the best working state.

[0112] Receive and input the original prompt word (S202):

[0113] The system receives raw prompts from the user in real time and inputs these prompts into the pre-configured generative model.

[0114] Translate the original prompt (S203):

[0115] Generative models translate the original input prompts based on the target language requirements set by the user. The translation process may involve steps such as language model decoding and word order adjustment to ensure the accuracy and fluency of the translation results.

[0116] Extract the text data information from the translation results (S204):

[0117] The system extracts text data information from the translation results, which forms the basis for subsequent processing of synonyms or near-synonyms.

[0118] Identify words in text data (S205):

[0119] The system identifies words in the extracted text data and obtains all contained words to prepare for subsequent processing of synonyms or near-synonyms.

[0120] Processing synonyms or near-synonyms (S206):

[0121] For each identified word, the system searches for its corresponding synonyms or near-synonyms. This can be achieved by querying a thesaurus or using semantic similarity algorithms.

[0122] Replace and generate multiple versions of text data information (S207):

[0123] The system replaces the original words with the found synonyms or near-synonyms in turn, generating multiple versions of text data. These versions, while retaining the original meaning, provide different expressions, increasing the diversity and richness of the text.

[0124] The above technical solution achieves the following effects: Through synonym or near-synonym processing, the system can generate multiple versions of text data, providing users with diverse expression options to meet the text generation requirements of different scenarios and needs. The replacement of synonyms or near-synonyms makes the text expression more vivid and rich, improving readability and attractiveness, and enhancing the user experience. Through the translation function of the generative model, the system can accurately translate the original prompts into the target language, while the combination of synonym or near-synonym processing improves the flexibility and accuracy of the translation. This technical solution reduces the workload of manual intervention and editing through automated processing, improving the efficiency of text generation. Simultaneously, through refined synonym or near-synonym processing, the effect of text generation is optimized, making the generated text more in line with user expectations and requirements.

[0125] In summary, this technical solution achieves the translation of the original prompts input by the user and the processing of synonyms or near-synonyms through a series of steps, generating multiple versions of text data information, providing diverse text expression options, improving the accuracy and flexibility of text translation, and optimizing the efficiency and effect of text generation.

[0126] In one embodiment of the present invention, synonym or near-synonym processing is performed on the words contained in the text data information to generate near-synonyms or near-synonyms corresponding to the words, including:

[0127] S2061. Perform semantic analysis on the words contained in the text data information to obtain the word meaning and context corresponding to the words;

[0128] S2062. Obtain the synonyms or near-synonyms corresponding to the words based on the meaning and context of the words;

[0129] S2063. Generate a list of synonyms or near-synonyms corresponding to the word.

[0130] The working principle of the above technical solution is: semantic analysis and contextual understanding (S2061):

[0131] Lexical analysis is performed on each word in the text data, that is, analyzing the basic meaning of the word and its specific meaning in different contexts. This usually involves analyzing the semantics, part of speech, and usage of the words.

[0132] At the same time, the system will also understand the context in which the words are located, that is, the position of the words in the sentence or paragraph, their relationship with other words, and the theme or emotional tendency of the overall text.

[0133] Get synonyms or near-synonyms (S2062):

[0134] Based on the results of semantic analysis and contextual understanding, the system uses synonym or near-synonym algorithms to find words with similar or identical meanings to the original words. These algorithms may be implemented based on methods such as word vectors, semantic similarity calculation, and thesaurus.

[0135] When acquiring synonyms or near-synonyms, the system also considers the usage habits and frequency of words in specific contexts to ensure that the found words not only have similar meanings, but also match the original words in terms of usage and context.

[0136] Generate a list of synonyms or near-synonyms (S2063):

[0137] The system compiles a list of synonyms or near-synonyms for each word found. This list contains all words that have similar or identical meanings to the original word.

[0138] The above list provides a wealth of options for subsequent synonym or near-synonym replacements, enabling the generated text data to have more ways of expression and style while maintaining the original meaning.

[0139] The effects of the above technical solution are as follows: By generating a list of synonyms or near-synonyms, the system can provide multiple alternatives for the original words, making the generated text data richer and more diverse in expression. This technical solution helps meet the personalized needs of different users or in different scenarios. The replacement of synonyms or near-synonyms can make the text expression more vivid, specific, and easier to understand. This technical solution helps improve the reader's reading experience and makes the text more attractive. Through semantic analysis and contextual understanding, the system can ensure that the found synonyms or near-synonyms are similar or identical in meaning to the original words, thereby optimizing the expression effect while maintaining the original meaning of the text. This technical solution reduces the workload of manual intervention and editing through automated processing, improving the efficiency of text generation. At the same time, by utilizing synonym or near-synonym algorithms, the system can intelligently select the most appropriate words for replacement, improving the level of intelligence in text generation.

[0140] In summary, this technical solution improves the richness, readability, and attractiveness of text expression by performing semantic analysis on words in text data, obtaining synonyms or near-synonyms, and generating a list, while optimizing the efficiency and intelligence level of text generation.

[0141] One embodiment of the present invention involves obtaining synonyms or near-synonyms for the word based on its meaning and context using a synonym or near-synonym algorithm, including:

[0142] Step 1: Use a pre-trained language model to represent the words using word vectors;

[0143] Step 2: Calculate the similarity between a word and other words in the thesaurus using word vectors; the similarity is obtained using the following formula:

[0144]

[0145] Where S represents the similarity score between the word and other words in the thesaurus / near-synonyms; A 01 A represents the word vector corresponding to a word; 02 represents the word vector corresponding to other words in the thesaurus / near-synonyms; 'a' represents the similarity compensation coefficient.

[0146]

[0147] Where n represents the dimension of the word vector corresponding to the word and the word vectors corresponding to other words in the synonym / near-synonym dictionary; x i This represents the i-th element of the word vector corresponding to the word; y i This represents the i-th element of the word vector corresponding to other words in the thesaurus / near-synonyms; c 0i This represents the preset reference data for the i-th vector element;

[0148] Step 3: Select words with a similarity greater than a preset similarity threshold as candidate synonyms / near-synonyms for the given word;

[0149] Step 4: Convert the candidate synonyms / near-synonyms to generate synonyms / near-synonyms corresponding to the words.

[0150] The working principle of the above technical solution is as follows: word vector representation (step 1):

[0151] Words are converted into high-dimensional vector representations, or word vectors, using pre-trained language models (such as Word2Vec, BERT, etc.). Word vectors can capture the semantic information of words, making semantically similar words close in distance in the vector space.

[0152] Similarity calculation (step 2):

[0153] Using word vectors, the similarity between a word and other words in a thesaurus is calculated using a similarity formula. A composite similarity calculation method based on cosine similarity and pre-defined vector element reference data is employed. Cosine similarity measures the angle between two vectors, while the pre-defined vector element reference data can be adjusted as needed to compensate for differences in different dimensions.

[0154] Candidate synonyms / near-synonyms selection (step 3):

[0155] Based on the similarity calculation results, words with a similarity score greater than a preset similarity threshold are selected as candidate synonyms / near-synonyms of the original words. The preset similarity threshold can be set according to actual needs to control the quantity and quality of candidate synonyms / near-synonyms.

[0156] Synonym / Near-synonym generation (Step 4):

[0157] The selected candidate synonyms / near-synonyms are transformed or processed as necessary, such as removing duplicates and sorting, to finally generate a list of synonyms / near-synonyms corresponding to the original words.

[0158] The above technical solution achieves the following results: By utilizing a pre-trained language model for word vector representation, it can capture the deep semantic information of words, thereby improving the accuracy of synonym / near-synonym searches. Simultaneously, through a composite similarity calculation method, it can comprehensively consider information from multiple dimensions, further enhancing the accuracy of search results. This technical solution can adjust the preset similarity threshold and preset vector element reference data according to actual needs to adapt to the synonym / near-synonym search requirements in different scenarios. This makes the technical solution highly flexible and customizable. Through automated word vector representation and similarity calculation processes, the workload of manually searching for synonyms / near-synonyms can be greatly reduced, improving the efficiency of text processing. The generated synonym / near-synonym list provides more choices for text expression, helping to enrich the content and form of the text and improve its readability and appeal.

[0159] In summary, this technical solution utilizes word vector representation and similarity calculation techniques to achieve accurate, flexible, and efficient synonym / near-synonym search and generation functions, providing strong support for text processing and applications.

[0160] One embodiment of the present invention involves real-time monitoring of user usage of synonyms or near-synonyms in multiple versions of text data, evaluating each synonym or near-synonym based on the usage, and determining the priority of the synonym or near-synonym based on the evaluation results, including:

[0161] S301. Monitor in real time the user's usage parameters for synonyms or near-synonyms in the text data information; wherein, the usage parameters include the number of times and frequency of use of synonyms or near-synonyms;

[0162] S302. Obtain the evaluation parameters of the synonym or near-synonym based on the usage parameters of the synonym or near-synonym; wherein, the evaluation parameters are obtained by the following formula:

[0163]

[0164] Where f represents the evaluation parameter corresponding to the synonym or near-synonym; δ represents the parameter baseline value; N represents the total number of times a synonym or near-synonym has been used; M represents the total number of times all words with the same meaning as a synonym or near-synonym have been used; n represents the number of unit fields entered by the user, and the number of characters contained in the unit field ranges from 300 to 500; ξ represents the first compensation coefficient; σ represents the second compensation coefficient;

[0165] The expression for the reference value δ is as follows:

[0166]

[0167] Among them, C i This represents the number of synonyms or near-synonyms contained in the i-th unit field; C ni This represents the number of times all words in the i-th unit field that have the same meaning as a synonym or near-synonym are used;

[0168] Furthermore, the expression for the first compensation coefficient ξ is as follows:

[0169]

[0170] Meanwhile, the expression for the second compensation coefficient σ is as follows:

[0171]

[0172] Among them, C i-1 This indicates the number of synonyms or near-synonyms contained in the (i-1)th unit field;

[0173] S303. Set the priority of the synonyms or near-synonyms according to the evaluation parameters from high to low, and generate the translation result according to the synonym or near-synonym with the highest priority when the user performs the next translation.

[0174] The working principle of the above technical solution is as follows: real-time monitoring of usage parameters (S301):

[0175] The system tracks users' usage of synonyms or near-synonyms in text data in real time, recording the number of times and frequency of use. These usage parameters reflect users' preferences and selection tendencies regarding synonyms or near-synonyms.

[0176] Obtain evaluation parameters (S302):

[0177] Based on the collected usage parameters, the system calculates the evaluation parameters for each synonym or near-synonym using a series of formulas. These formulas comprehensively consider factors such as the total number of times the synonym or near-synonym is used, the total number of times words with the same meaning are used, the number of unit fields, and the compensation coefficient. In this way, the system can quantify the importance and value of synonyms or near-synonyms.

[0178] The baseline parameter δ measures the usage of synonyms or near-synonyms within a unit field, reflecting the degree of user demand for different synonyms or near-synonyms. The first compensation coefficient ξ and the second compensation coefficient σ are used to adjust the evaluation parameters to better reflect actual usage.

[0179] Set priority and apply (S303):

[0180] Based on the evaluation parameters, the system assigns a priority to each synonym or near-synonym. This way, when the user performs a translation next time, the system will generate a translation based on the highest-priority synonym or near-synonym, thereby optimizing translation quality and user experience.

[0181] The effects of the above technical solution are as follows: By monitoring and evaluating the usage of synonyms or near-synonyms in real time, the system can provide users with translation results that better match their preferences and needs, thereby improving translation quality and user experience. Since the system can prioritize each synonym or near-synonym based on user usage, personalized translation can be achieved, meeting diverse user translation needs. By prioritizing the use of synonyms or near-synonyms with higher evaluation parameters, the system can reduce unnecessary selection processes during text generation, thus optimizing text generation efficiency. Through real-time monitoring, evaluation, and priority setting, this technical solution enables the system to more intelligently understand and respond to user needs, improving the system's intelligence level.

[0182] In summary, this technical solution achieves personalized and efficient translation services by monitoring and evaluating users' use of synonyms or near-synonyms in real time and setting priorities based on the evaluation results, thereby improving user experience and system intelligence.

[0183] This invention proposes a generative model-guided word testing system, such as... Figure 2 As shown, the generative model guide word testing system includes:

[0184] The parameter configuration module is used to configure the model parameters of the generative model, wherein the generative model adopts a model based on a large language model, such as the GPT series model.

[0185] The translation execution module is used to translate the original prompt words input by the user using the generative model, obtain translation results, and process the translation results for synonyms or near-synonyms to obtain multiple versions of text data information after processing.

[0186] The real-time monitoring module is used to monitor the user's usage of synonyms or near-synonyms in the multiple versions of text data information in real time, evaluate each synonym or near-synonym based on the usage, and determine the priority of the synonym or near-synonym based on the evaluation results.

[0187] The working principle of the above technical solution is as follows: First, configure the model parameters of the generative model. The generative model here is based on large language models, such as the GPT series models. These models are trained on a large amount of text data and are able to generate fluent and context-sensitive text.

[0188] Next, the configured generative model is used to translate the original prompts input by the user. The purpose of translation is to transform the original prompts into a form that is easier for the model to understand or generate.

[0189] After translation, synonyms or near-synonyms in the translation results are processed. This processing may include the identification, replacement, or expansion of synonyms or near-synonyms to generate multiple versions of the text data.

[0190] Real-time monitoring of user usage of synonyms or near-synonyms across multiple versions of text data. This can be achieved by tracking user interactions with the system and analyzing user behavior logs.

[0191] Each synonym or near-synonym is evaluated based on user behavior. The evaluation may be based on multiple dimensions, such as user selection frequency, usage duration, and feedback.

[0192] Based on the evaluation results, the priorities of synonyms or near-synonyms are determined. Synonyms or near-synonyms with higher priorities are more likely to be recommended to users by the system, thereby improving user satisfaction and system accuracy.

[0193] The effects of the above technical solution are as follows: By generating multiple versions of text data and providing the option to choose synonyms or near-synonyms, the system can better meet users' personalized needs and improve user experience. By monitoring and evaluating user behavior in real time, the system can understand users' preferences for synonyms or near-synonyms, thereby recommending appropriate words more accurately and improving the accuracy and relevance of the generated text. By collecting and analyzing user feedback and behavioral data, the parameters and algorithms of the generative model can be continuously optimized, improving the model's performance and adaptability. By continuously learning and adapting to users' usage habits, the system can gradually improve its intelligence, providing users with more intelligent and personalized services.

[0194] In summary, this technical solution, by configuring generative models, translating prompts, handling synonyms or near-synonyms, and monitoring and evaluating user behavior in real time, can improve user experience, increase accuracy, optimize model performance, and enhance system intelligence.

[0195] In one embodiment of the present invention, the translation execution module includes:

[0196] An initialization module is used to initialize the generative model after the parameter configuration is completed;

[0197] The data input module is used to receive the original prompt words input by the user in real time and input the received original prompt words into the generative model with completed parameter configuration;

[0198] The translation control module is used to translate the original prompt words according to the user's target language requirements through the generative model with completed parameter configuration, and obtain the translation result;

[0199] The text data information extraction module is used to extract the text data information corresponding to the translation result;

[0200] The word acquisition module is used to identify words in the text data information corresponding to the translation result and acquire the words contained in the text data information;

[0201] The synonym / near-synonym processing module is used to process the words contained in the text data information into near-synonyms or near-synonyms, and generate near-synonyms or near-synonyms corresponding to the words;

[0202] The replacement operation module is used to sequentially replace the corresponding words with the synonyms or near-synonyms, generating multiple versions of text data information with synonyms or near-synonyms corresponding to the text data information.

[0203] The working principle of the above technical solution is as follows: before starting the translation, the generative model with completed parameter configuration is first initialized to ensure that the model is in the best working state.

[0204] The system receives raw prompts from the user in real time and inputs these prompts into the pre-configured generative model.

[0205] Generative models translate the original input prompts based on the target language requirements set by the user. The translation process may involve steps such as language model decoding and word order adjustment to ensure the accuracy and fluency of the translation results.

[0206] The system extracts text data information from the translation results, which forms the basis for subsequent processing of synonyms or near-synonyms.

[0207] The system identifies words in the extracted text data and obtains all contained words to prepare for subsequent processing of synonyms or near-synonyms.

[0208] For each identified word, the system searches for its corresponding synonyms or near-synonyms. This can be achieved by querying a thesaurus or using semantic similarity algorithms.

[0209] The system replaces the original words with the found synonyms or near-synonyms in turn, generating multiple versions of text data. These versions, while retaining the original meaning, provide different expressions, increasing the diversity and richness of the text.

[0210] The above technical solution achieves the following effects: Through synonym or near-synonym processing, the system can generate multiple versions of text data, providing users with diverse expression options to meet the text generation requirements of different scenarios and needs. The replacement of synonyms or near-synonyms makes the text expression more vivid and rich, improving readability and attractiveness, and enhancing the user experience. Through the translation function of the generative model, the system can accurately translate the original prompts into the target language, while the combination of synonym or near-synonym processing improves the flexibility and accuracy of the translation. This technical solution reduces the workload of manual intervention and editing through automated processing, improving the efficiency of text generation. Simultaneously, through refined synonym or near-synonym processing, the effect of text generation is optimized, making the generated text more in line with user expectations and requirements.

[0211] In summary, this technical solution achieves the translation of the original prompts input by the user and the processing of synonyms or near-synonyms through a series of steps, generating multiple versions of text data information, providing diverse text expression options, improving the accuracy and flexibility of text translation, and optimizing the efficiency and effect of text generation.

[0212] In one embodiment of the present invention, the synonym / near-synonym processing module includes:

[0213] The word meaning analysis module is used to perform word meaning analysis on the words contained in the text data information to obtain the word meaning and context corresponding to the words;

[0214] The synonym or near-synonym acquisition module is used to acquire the synonyms or near-synonyms corresponding to the words based on the meaning and context of the words;

[0215] The list generation module is used to generate a list of synonyms or near-synonyms corresponding to the word.

[0216] The working principle of the above technical solution is as follows: semantic analysis is performed on each word in the text data, that is, the basic meaning of the word and its specific meaning in different contexts are analyzed. This usually involves the analysis of the semantics, part of speech, and usage of the words.

[0217] At the same time, the system will also understand the context in which the words are located, that is, the position of the words in the sentence or paragraph, their relationship with other words, and the theme or emotional tendency of the overall text.

[0218] Based on the results of semantic analysis and contextual understanding, the system uses synonym or near-synonym algorithms to find words with similar or identical meanings to the original words. These algorithms may be implemented based on methods such as word vectors, semantic similarity calculation, and thesaurus.

[0219] When acquiring synonyms or near-synonyms, the system also considers the usage habits and frequency of words in specific contexts to ensure that the found words not only have similar meanings, but also match the original words in terms of usage and context.

[0220] The system compiles a list of synonyms or near-synonyms for each word found. This list contains all words that have similar or identical meanings to the original word.

[0221] The above list provides a wealth of options for subsequent synonym or near-synonym replacements, enabling the generated text data to have more ways of expression and style while maintaining the original meaning.

[0222] The effects of the above technical solution are as follows: By generating a list of synonyms or near-synonyms, the system can provide multiple alternatives for the original words, making the generated text data richer and more diverse in expression. This technical solution helps meet the personalized needs of different users or in different scenarios. The replacement of synonyms or near-synonyms can make the text expression more vivid, specific, and easier to understand. This technical solution helps improve the reader's reading experience and makes the text more attractive. Through semantic analysis and contextual understanding, the system can ensure that the found synonyms or near-synonyms are similar or identical in meaning to the original words, thereby optimizing the expression effect while maintaining the original meaning of the text. This technical solution reduces the workload of manual intervention and editing through automated processing, improving the efficiency of text generation. At the same time, by utilizing synonym or near-synonym algorithms, the system can intelligently select the most appropriate words for replacement, improving the level of intelligence in text generation.

[0223] In summary, this technical solution improves the richness, readability, and attractiveness of text expression by performing semantic analysis on words in text data, obtaining synonyms or near-synonyms, and generating a list, while optimizing the efficiency and intelligence level of text generation.

[0224] In one embodiment of the present invention, the synonym or near-synonym acquisition module includes:

[0225] The word vector representation module is used to represent the words using a pre-trained language model;

[0226] The similarity acquisition module is used to calculate the similarity between a word and other words in the thesaurus / near-synonyms using word vector representation; wherein, the similarity is obtained by the following formula:

[0227]

[0228] Where S represents the similarity score between the word and other words in the thesaurus / near-synonyms; A 01 A represents the word vector corresponding to a word; 02 represents the word vector corresponding to other words in the thesaurus / near-synonyms; 'a' represents the similarity compensation coefficient.

[0229]

[0230] Where n represents the dimension of the word vector corresponding to the word and the word vectors corresponding to other words in the synonym / near-synonym dictionary; x i This represents the i-th element of the word vector corresponding to the word; y i This represents the i-th element of the word vector corresponding to other words in the thesaurus / near-synonyms; c 0i This represents the preset reference data for the i-th vector element;

[0231] The candidate synonym / near-synonym acquisition module is used to select words with a similarity greater than a preset similarity threshold as candidate synonyms / near-synonyms corresponding to the word.

[0232] The synonym / near-synonym generation module is used to convert the candidate synonyms / near-synonyms to generate synonyms / near-synonyms corresponding to the words.

[0233] The working principle of the above technical solution is as follows: words are converted into high-dimensional vector representations, i.e., word vectors, using pre-trained language models (such as Word2Vec, BERT, etc.). Word vectors can capture the semantic information of words, so that semantically similar words are also close in distance in the vector space.

[0234] Using word vectors, the similarity between a word and other words in a thesaurus is calculated using a similarity formula. A composite similarity calculation method based on cosine similarity and pre-defined vector element reference data is employed. Cosine similarity measures the angle between two vectors, while the pre-defined vector element reference data can be adjusted as needed to compensate for differences in different dimensions.

[0235] Based on the similarity calculation results, words with a similarity score greater than a preset similarity threshold are selected as candidate synonyms / near-synonyms of the original words. The preset similarity threshold can be set according to actual needs to control the quantity and quality of candidate synonyms / near-synonyms.

[0236] The selected candidate synonyms / near-synonyms are transformed or processed as necessary, such as removing duplicates and sorting, to finally generate a list of synonyms / near-synonyms corresponding to the original words.

[0237] The above technical solution achieves the following results: By utilizing a pre-trained language model for word vector representation, it can capture the deep semantic information of words, thereby improving the accuracy of synonym / near-synonym searches. Simultaneously, through a composite similarity calculation method, it can comprehensively consider information from multiple dimensions, further enhancing the accuracy of search results. This technical solution can adjust the preset similarity threshold and preset vector element reference data according to actual needs to adapt to the synonym / near-synonym search requirements in different scenarios. This makes the technical solution highly flexible and customizable. Through automated word vector representation and similarity calculation processes, the workload of manually searching for synonyms / near-synonyms can be greatly reduced, improving the efficiency of text processing. The generated synonym / near-synonym list provides more choices for text expression, helping to enrich the content and form of the text and improve its readability and appeal.

[0238] In summary, this technical solution utilizes word vector representation and similarity calculation techniques to achieve accurate, flexible, and efficient synonym / near-synonym search and generation functions, providing strong support for text processing and applications.

[0239] In one embodiment of the present invention, the real-time monitoring module includes:

[0240] A real-time parameter monitoring module is used to monitor the user's usage parameters for synonyms or near-synonyms in the text data information in real time; wherein, the usage parameters include the number of times and frequency of use of synonyms or near-synonyms;

[0241] The evaluation parameter acquisition module is used to acquire the evaluation parameters of the synonyms or near-synonyms based on their usage parameters; wherein the evaluation parameters are acquired using the following formula:

[0242]

[0243] Where f represents the evaluation parameter corresponding to the synonym or near-synonym; δ represents the parameter baseline value; N represents the total number of times a synonym or near-synonym has been used; M represents the total number of times all words with the same meaning as a synonym or near-synonym have been used; n represents the number of unit fields entered by the user, and the number of characters contained in the unit field ranges from 300 to 500; ξ represents the first compensation coefficient; σ represents the second compensation coefficient;

[0244] The expression for the reference value δ is as follows:

[0245]

[0246] Among them, C i This represents the number of synonyms or near-synonyms contained in the i-th unit field; C ni This represents the number of times all words in the i-th unit field that have the same meaning as a synonym or near-synonym are used;

[0247] Furthermore, the expression for the first compensation coefficient ξ is as follows:

[0248]

[0249] Meanwhile, the expression for the second compensation coefficient σ is as follows:

[0250]

[0251] Among them, C i-1 This indicates the number of synonyms or near-synonyms contained in the (i-1)th unit field;

[0252] The priority processing module is used to set the priority of the synonyms or near-synonyms in descending order of the evaluation parameters, and generate the translation result according to the synonym or near-synonym with the highest priority when the user performs the next translation.

[0253] The working principle of the above technical solution is as follows: the system tracks the user's usage of synonyms or near-synonyms in text data in real time, recording their usage frequency and number of times. These usage parameters can reflect the user's preference and selection tendency for synonyms or near-synonyms.

[0254] Based on the collected usage parameters, the system calculates the evaluation parameters for each synonym or near-synonym using a series of formulas. These formulas comprehensively consider factors such as the total number of times the synonym or near-synonym is used, the total number of times words with the same meaning are used, the number of unit fields, and the compensation coefficient. In this way, the system can quantify the importance and value of synonyms or near-synonyms.

[0255] The baseline parameter δ measures the usage of synonyms or near-synonyms within a unit field, reflecting the degree of user demand for different synonyms or near-synonyms. The first compensation coefficient ξ and the second compensation coefficient σ are used to adjust the evaluation parameters to better reflect actual usage.

[0256] Based on the evaluation parameters, the system assigns a priority to each synonym or near-synonym. This way, when the user performs a translation next time, the system will generate a translation based on the highest-priority synonym or near-synonym, thereby optimizing translation quality and user experience.

[0257] The effects of the above technical solution are as follows: By monitoring and evaluating the usage of synonyms or near-synonyms in real time, the system can provide users with translation results that better match their preferences and needs, thereby improving translation quality and user experience. Since the system can prioritize each synonym or near-synonym based on user usage, personalized translation can be achieved, meeting diverse user translation needs. By prioritizing the use of synonyms or near-synonyms with higher evaluation parameters, the system can reduce unnecessary selection processes during text generation, thus optimizing text generation efficiency. Through real-time monitoring, evaluation, and priority setting, this technical solution enables the system to more intelligently understand and respond to user needs, improving the system's intelligence level.

[0258] In summary, this technical solution achieves personalized and efficient translation services by monitoring and evaluating users' use of synonyms or near-synonyms in real time and setting priorities based on the evaluation results, thereby improving user experience and system intelligence.

[0259] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A generative model guide word testing method, characterized in that, The generative model guide word testing method includes: Configure the model parameters for the generative model; The generative model is used to translate the original prompts input by the user to obtain translation results, and the translation results are processed by synonyms or near-synonyms to obtain multiple versions of text data information after processing. Real-time monitoring of user usage of synonyms or near-synonyms in multiple versions of text data, evaluation of each synonym or near-synonym based on the usage, and determination of the priority of the synonym or near-synonym based on the evaluation results, including: The system monitors in real time the user's usage parameters for synonyms or near-synonyms in the text data; wherein, the usage parameters include the number of times and frequency of use of synonyms or near-synonyms. The evaluation parameters of the synonyms or near-synonyms are obtained based on their usage parameters; wherein, the evaluation parameters are obtained using the following formula: ； in, f This represents the evaluation parameters corresponding to the synonyms or near-synonyms; δ Indicates the baseline value of the parameter; N This indicates the total number of times a synonym or near-synonym has been used. M This indicates the total number of times all words with the same meaning as a synonym or near-synonym are used; n This indicates the number of unit fields that the user has entered, and the number of characters contained in the unit field ranges from 300 to 500. ξ Indicates the first compensation coefficient; σ Indicates the second compensation coefficient; Wherein, the parameter reference value δ The expression is as follows: ； in, C i Indicates the first i The number of synonyms or near-synonyms contained in each unit field; C ni Indicates the first i The number of times each unit field contains all words that have the same meaning as a synonym or near-synonym are used; Furthermore, the first compensation coefficient ξ The expression is as follows: ； Meanwhile, the second compensation coefficient σ The expression is as follows: ； in, C i-1 Indicates the first i -1 The number of synonyms or near-synonyms contained in a single unit field; The priorities of the synonyms or near-synonyms are set according to the evaluation parameters from high to low, and the translation result is generated according to the synonym or near-synonym with the highest priority when the user performs the next translation.

2. The generative model guide word testing method according to claim 1, characterized in that, The generative model is used to translate the original prompts input by the user, obtaining translation results. The translation results are then processed using synonyms or near-synonyms to obtain multiple versions of processed text data, including: Initialize the generative model with the parameters configured. The system receives raw prompts from the user in real time and inputs them into the generative model with the parameters configured. The generative model with the completed parameter configuration translates the original prompt words according to the user's target language requirements to obtain the translation result; Extract the text data information corresponding to the translation result; The words in the text data information corresponding to the translation result are identified to obtain the words contained in the text data information; The words contained in the text data are processed to generate synonyms or near-synonyms corresponding to the words. The synonyms or near-synonyms are replaced sequentially with their corresponding words to generate multiple versions of text data information with synonyms or near-synonyms that correspond to the text data information.

3. The generative model guide word testing method according to claim 2, characterized in that, The text data information is processed to generate synonyms or near-synonyms for the words, including: Perform semantic analysis on the words contained in the text data to obtain the word meaning and context corresponding to the words; Based on the meaning and context of the word, synonyms or near-synonyms are obtained using a synonym or near-synonym algorithm. Generate a list of synonyms or near-synonyms corresponding to the given word.

4. The generative model guide word testing method according to claim 3, characterized in that, Based on the meaning and context of the word, synonyms or near-synonyms are obtained using a synonym or near-synonym algorithm, including: The words are represented by word vectors using a pre-trained language model; Word vectors are used to calculate the similarity between a word and other words in a thesaurus / near-synonyms; wherein, the similarity is obtained by the following formula: ； in, S This represents the numerical similarity score between the stated word and other words in the thesaurus / near-synonyms. A 01 This represents the word vector corresponding to a word; A 02 This represents the word vectors corresponding to other words in the thesaurus / near-synonyms; a This represents the similarity compensation coefficient; ； in, n This represents the dimension of the word vector corresponding to the word and the word vectors corresponding to other words in the thesaurus / near-synonyms; x i The word vector corresponding to a word is the first... i One element; y i This represents the first word vector corresponding to other words in the thesaurus / near-synonyms. i One element; c 0i Indicates the preset first i Each vector element references data; Words with a similarity greater than a preset similarity threshold are selected as candidate synonyms / near-synonyms for the given word; The candidate synonyms / near-synonyms are converted to generate synonyms / near-synonyms corresponding to the words.

5. A generative model-guided word testing system, characterized in that, The generative model-guided word testing system includes: The parameter configuration module is used to configure the model parameters of the generative model; The translation execution module is used to translate the original prompt words input by the user using the generative model, obtain translation results, and process the translation results for synonyms or near-synonyms to obtain multiple versions of text data information after processing. The real-time monitoring module is used to monitor the user's usage of synonyms or near-synonyms in the multiple versions of text data information in real time, evaluate each synonym or near-synonym based on the usage, and determine the priority of the synonym or near-synonym based on the evaluation results; The real-time monitoring module includes: A real-time parameter monitoring module is used to monitor the user's usage parameters for synonyms or near-synonyms in the text data information in real time; wherein, the usage parameters include the number of times and frequency of use of synonyms or near-synonyms; The evaluation parameter acquisition module is used to acquire the evaluation parameters of the synonyms or near-synonyms based on their usage parameters; wherein the evaluation parameters are acquired using the following formula: ； in, f This represents the evaluation parameters corresponding to the synonyms or near-synonyms; δ Indicates the baseline value of the parameter; N This indicates the total number of times a synonym or near-synonym has been used. M This indicates the total number of times all words with the same meaning as a synonym or near-synonym are used; n This indicates the number of unit fields that the user has entered, and the number of characters contained in the unit field ranges from 300 to 500. ξ Indicates the first compensation coefficient; σ Indicates the second compensation coefficient; Wherein, the parameter reference value δ The expression is as follows: ； in, C i Indicates the first i The number of synonyms or near-synonyms contained in each unit field; C ni Indicates the first i The number of times each unit field contains all words that have the same meaning as a synonym or near-synonym are used; Furthermore, the first compensation coefficient ξ The expression is as follows: ； Meanwhile, the second compensation coefficient σ The expression is as follows: ； in, C i-1 Indicates the first i -1 The number of synonyms or near-synonyms contained in a single unit field; The priority processing module is used to set the priority of the synonyms or near-synonyms in descending order of the evaluation parameters, and generate the translation result according to the synonym or near-synonym with the highest priority when the user performs the next translation.

6. The generative model guide word testing system according to claim 5, characterized in that, The translation execution module includes: An initialization module is used to initialize the generative model after the parameter configuration is completed; The data input module is used to receive the original prompt words input by the user in real time and input the received original prompt words into the generative model with completed parameter configuration; The translation control module is used to translate the original prompt words according to the user's target language requirements through the generative model with completed parameter configuration, and obtain the translation result; The text data information extraction module is used to extract the text data information corresponding to the translation result; The word acquisition module is used to identify words in the text data information corresponding to the translation result and acquire the words contained in the text data information; The synonym / near-synonym processing module is used to process the words contained in the text data information into near-synonyms or near-synonyms, and generate near-synonyms or near-synonyms corresponding to the words; The replacement operation module is used to sequentially replace the corresponding words with the synonyms or near-synonyms, generating multiple versions of text data information with synonyms or near-synonyms corresponding to the text data information.

7. The generative model guide word testing system according to claim 6, characterized in that, The synonym / near-synonym processing module includes: The word meaning analysis module is used to perform word meaning analysis on the words contained in the text data information to obtain the word meaning and context corresponding to the words; The synonym or near-synonym acquisition module is used to acquire the synonyms or near-synonyms corresponding to the words based on the meaning and context of the words; The list generation module is used to generate a list of synonyms or near-synonyms corresponding to the word.

8. The generative model guide word testing system according to claim 7, characterized in that, The synonym or near-synonym acquisition module includes: The word vector representation module is used to represent the words using a pre-trained language model; The similarity acquisition module is used to calculate the similarity between a word and other words in the thesaurus / near-synonyms using word vector representation; wherein, the similarity is obtained by the following formula: ； in, S This represents the numerical similarity score between the stated word and other words in the thesaurus / near-synonyms. A 01 This represents the word vector corresponding to a word; A 02 This represents the word vectors corresponding to other words in the thesaurus / near-synonyms; a This represents the similarity compensation coefficient; ； in, n This represents the dimension of the word vector corresponding to the word and the word vectors corresponding to other words in the thesaurus / near-synonyms; x i The word vector corresponding to a word is the first... i One element; y i This represents the first word vector corresponding to other words in the thesaurus / near-synonyms. i One element; c 0i Indicates the preset first i Each vector element references data; The candidate synonym / near-synonym acquisition module is used to select words with a similarity greater than a preset similarity threshold as candidate synonyms / near-synonyms corresponding to the word. The synonym / near-synonym generation module is used to convert the candidate synonyms / near-synonyms to generate synonyms / near-synonyms corresponding to the words.

Images