CN118798201B - New proposition error correction method and system based on large model - Google Patents
New proposition error correction method and system based on large model Download PDFInfo
- Publication number
- CN118798201B CN118798201B CN202411272928.5A CN202411272928A CN118798201B CN 118798201 B CN118798201 B CN 118798201B CN 202411272928 A CN202411272928 A CN 202411272928A CN 118798201 B CN118798201 B CN 118798201B
- Authority
- CN
- China
- Prior art keywords
- new
- proposition
- new proposition
- propositions
- calculating
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
- G06F40/295—Named entity recognition
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/253—Grammatical analysis; Style critique
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
- G06N5/025—Extracting rules from data
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Animal Behavior & Ethology (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a new proposition error correction method and system based on a big model, wherein the method comprises the steps of S1, obtaining the generated new proposition and corresponding context information thereof, S2, carrying out semantic analysis on the new proposition based on a pre-trained language big model, identifying potential errors in the new proposition, S3, verifying the accuracy of the new proposition by using a domain-specific knowledge base, S4, automatically correcting errors in the new proposition according to analysis results and verification feedback, and the new proposition error correction method and system based on the big model solve the problems of how to develop a more intelligent and flexible error correction method to improve the accuracy and efficiency when processing the new proposition, and still has the problems of insufficient identification accuracy, limited processing capacity on the new proposition errors and the like although the prior art has some researches on attempting to improve the error correction effect by enhancing the adaptability of the model or combining various technical means.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a new proposition error correction method and system based on a large model.
Background
In the prior art, automatic error correction techniques based on Natural Language Processing (NLP) and machine learning algorithms have been widely used, which typically recognize and correct grammatical, spelling and logic errors in text by training models, particularly with significant progress in grammar correction and spell checking. Existing error correction systems often rely on predefined rules or dictionaries, in combination with statistical or deep learning models to detect and correct errors, which techniques exhibit high accuracy and efficiency in processing standardized text, and with the development of natural language processing techniques, particularly the application of Large Language Models (LLMs), more complex and innovative text generation tasks have emerged. Large language models are able to generate text containing new propositions or non-standard expressions, which presents new challenges to existing error correction methods.
Existing error correction techniques typically rely on predefined rules or standard language constructs, but in the face of innovative representations of large model generation, these techniques tend to be difficult to adapt to and fail to effectively identify and correct errors therein. Especially when dealing with new propositions, existing error correction techniques tend to suffer from a high rate of false positives, as these expressions may not conform to existing language specifications or contain innovative logic deductions. Furthermore, the uniqueness and diversity of the new proposition makes it difficult for the prior art to cover all possible error types, which further increases the difficulty of error correction. Therefore, how to develop a more intelligent and flexible error correction method to improve accuracy and efficiency in processing new propositions becomes a key problem in current researches, and at present, although some researches have been attempted to improve the error correction effect by enhancing the adaptability of a model or combining various technical means, there are still problems of insufficient recognition accuracy, limited processing ability on new types of errors, and the like.
Disclosure of Invention
The invention aims to provide a new proposition error correction method and system based on a large model, which solve the problems of how to develop a more intelligent and flexible error correction method so as to improve the accuracy and efficiency when processing new propositions, and although some researches in the prior art attempt to improve the error correction effect by enhancing the adaptability of the model or combining a plurality of technical means, the problems of insufficient identification accuracy, limited processing capacity on new types of errors and the like still exist.
In order to achieve the purpose, the invention provides the following technical scheme that the new proposition error correction method based on the large model comprises the following steps:
S1, acquiring generated new propositions and corresponding context information thereof;
s2, carrying out semantic analysis on the new propositions based on the pre-trained language big model, and identifying potential errors in the new propositions;
s3, verifying the accuracy of the new proposition by using a domain specific knowledge base;
s4, automatically correcting errors in the new propositions according to the analysis result and verification feedback;
the step S1 of obtaining the generated new proposition and the corresponding context information thereof specifically comprises the following steps:
Receiving the output new proposition text through an API interface;
Extracting key words in the new proposition text;
Building a context based on the key words;
identifying entities and relationships in the new proposition using natural language processing techniques;
the calculating the similarity S between the new proposition and the standard expression in the field based on the semantic vector representation specifically comprises the following steps:
normalizing the semantic vector representation of the new proposition;
Calculating cosine similarity C between the normalized semantic vector representation and the intra-domain standard expression vector representation;
If the cosine similarity C is smaller than 0.7, the new proposition is not matched with the standard expression, namely when C is smaller than 0.7, the new proposition is judged to have larger difference from the standard expression;
Adjusting the calculation mode of the similarity S based on the mismatch condition;
the concrete steps of calculating the cosine similarity C between the normalized semantic vector representation and the standard expression vector representation in the field include:
determining a new propositional semantic vector representation V1 and a standard expression vector representation V2;
Calculating the dot product P of V1 and V2;
Calculating the module lengths M1 and M2 of V1 and V2;
calculating cosine similarity according to the formula cosine similarity c=p/(m1×m2);
the step S3 specifically includes:
Extracting relevant data from a predefined domain-specific knowledge base containing facts, rules, logical relationships, and domain-specific concepts;
let the knowledge base be K, the new topic be P,
For each sub-proposition Pi in P, find from K the set of propositions { K1, K2, & gt, kn } that is closest to Pi semantics,
Calculating the similarity of corresponding propositions Kj in propositions Pi and KThe cosine similarity formula is used:,
Wherein, AndIs a semantic vector representation of the corresponding proposition;
Verifying the logic consistency of the new proposition and the known knowledge according to rules and logic in a knowledge base, evaluating the overall accuracy of the new proposition PPP, comprehensively considering the similarity and logic consistency of all sub propositions, and obtaining the confidence score C (P) of the new proposition:
,
Wherein, Is a logical consistency score.
Preferably, the step of performing semantic analysis on the new proposition based on the pre-trained language big model in the step S2 specifically includes:
inputting the new proposition to the pre-trained language big model to obtain a semantic vector representation;
calculating the similarity S between the new proposition and the standard expression in the field based on the semantic vector representation;
if the similarity S is lower than a preset threshold T, determining that the new proposition possibly has semantic errors;
and comparing the new propositions with key difference points of standard expressions in the field.
Preferably, in the step S2, the semantic analysis is performed on the new proposition based on the pre-trained language big model, and the specific steps of the recognition process for recognizing the potential errors in the new proposition include:
acquiring lexical structure information of new propositions;
Calculating a grammar probability score P (grammar) of the new proposition based on a language model;
Judging whether the grammar probability score P (grammar) is lower than a first preset threshold value theta 1 or not;
If P (grammar) < θ1, identifying that a new proposition may have a grammar error;
The specific steps of calculating the grammar probability score P (grammar) of the new proposition based on the language model comprise the following steps:
Word segmentation processing is carried out on the new proposition to obtain a vocabulary sequence;
calculating a probability score P (vocabulary) of the vocabulary sequence based on the language model;
calculating a context correlation score P (context) for the new proposition;
the grammar probability score P (grammar) is calculated according to the formula P (grammar) =p (vocabulary) ×p (context).
Preferably, the specific step of calculating the context correlation score P (context) of the new proposition includes:
acquiring the context information of the new proposition;
Calculating relevance scores P (front) and P (rear) of the context information based on the language model;
judging whether the correlation scores P (front) and P (rear) are higher than a second preset threshold value theta 2 or not;
If P (front) > θ2 and P (rear) > θ2, then P (context) = (P (front) +p (rear))/2;
wherein, the specific steps of calculating the relevance scores P (front) and P (back) of the context information based on the language model comprise:
Extracting the context keywords of the new proposition;
Calculating the probability P (key words) of the key words occurring in the context based on a language model;
calculating semantic similarity S (semantics) between the keywords and the new propositions;
the correlation scores P (front) and P (rear) are calculated according to the formula P (front)/P (rear) =p (keyword) ×s (semantic).
Preferably, the specific step of calculating the semantic similarity S (semantic) between the keyword and the new proposition includes:
the keyword and the new proposition are respectively and vectorized to be represented as a vector V keyword and a vector V new proposition;
calculating cosine similarity COS between the vector V key words and the V new proposition;
If COS > the third preset threshold θ3, S (semantic) =cos.
Preferably, the step S4 specifically includes:
determining the position of the error in the proposition based on the similarity and the logical consistency result in the step S3, locating the error point of each sub-proposition Pi if Less than a preset thresholdThen consider that the sub-proposition has an error;
Error correction is carried out by utilizing the correct proposition sets { k1, k2, & gt, kn } in the domain-specific knowledge base, a replacement strategy is used for selecting a proposition Kj with highest similarity with Pi and best logic consistency to replace an error part in the Pi, and the proposition after correction is carried out The method comprises the following steps:
;
New propositions to be corrected Again through knowledge base verification, ensuring the consistency of the semantics and logic, and if the revised proposition still does not meet the conditions, further adjusting until the confidence level of the propositionReaching the set threshold value.
The new propositional error correction system based on the large model adopts the new propositional error correction method based on the large model, and the system comprises:
the context information acquisition module is used for acquiring new propositions generated by the large model and the corresponding context information;
the semantic analysis module is used for carrying out semantic analysis on the new proposition based on the pre-trained language model and identifying potential errors in the new proposition;
The verification module utilizes the domain-specific knowledge base to verify the accuracy of the new proposition;
and the correction module automatically corrects errors in the new propositions according to the semantic analysis result and verification feedback.
According to the technical scheme, the invention has the following beneficial effects:
According to the new proposition error correction method and system based on the large model, the generated new proposition and the corresponding context information thereof are acquired, semantic analysis is carried out on the new proposition based on the pre-trained language large model, potential errors in the new proposition are identified, the accuracy of the new proposition is verified by utilizing the domain-specific knowledge base, errors in the new proposition are automatically corrected according to analysis results and verification feedback, so that the new proposition is corrected, especially when new types of errors which are not clearly regular and can be circulated are processed, the new proposition error correction method is more intelligent and flexible, the accuracy and efficiency are improved, the problems that how to develop the more intelligent and flexible error correction method to improve the accuracy and efficiency when the new proposition is processed are solved, and although the prior art has some research attempts to improve the error correction effect through the adaptability of the enhanced model or combining various technical means, the problems that the identification accuracy is insufficient, the processing capacity on the new types of errors is limited and the like still exist.
Drawings
FIG. 1 is a schematic flow chart of the method of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, a new proposition error correction method based on a large model, the method includes:
S1, by acquiring the generated new proposition and the corresponding context information, the meaning of the proposition in a specific context can be comprehensively understood. This process typically involves extracting specific sentences or phrases from the output and collecting background information, conversation history, etc. associated therewith. For example, assume that a new proposition regarding climate change is generated that "the global air temperature has fallen 5 degrees in the past decade. To better understand this proposition we need to collect details about time frames, data sources, measurement criteria, etc. This step is critical for subsequent semantic analysis, as it provides the necessary context to assess the authenticity and accuracy of the proposition. In addition, in practical applications, it may be necessary to consider how to efficiently screen out relevant pieces of information from a large amount of text data to support subsequent analysis work.
S2, carrying out semantic analysis on the new proposition based on the pre-trained language big model to identify potential errors in the new proposition. At this stage, we use advanced natural language processing techniques, such as the pretrained models of BERT, GPT-3, etc., to analyze the semantic structure of the new proposition in depth. These models are trained on large amounts of text data to capture complex language patterns and context. For example, for the aforementioned new proposition regarding global air temperature changes, the pre-trained model may find that the expression "5 degrees down in the last decade" does not correspond to the known trend of climate change, because the global air temperature is actually rising rather than falling according to the existing scientific consensus. Such analysis is not limited to actual errors, but may also cover problems of logical inconsistencies, semantic ambiguity, etc. In this way we can more accurately identify potential problems in the new proposition.
S3, verifying the accuracy of the new proposition by using the domain-specific knowledge base, and further ensuring the reliability and the authenticity of the new proposition. The domain-specific knowledge base contains domain-specific expertise and data, such as a meteorological database, scientific research results, and the like. Continuing with the example of climate change above, we can query the data records in the meteorological database for global air temperature changes to verify that the statement "5 degrees down in the past decade" is correct. If the database shows that the global air temperature is actually rising, then it can be determined that the original proposition is wrong. In addition, expert systems or specialized literature can be utilized to further confirm the accuracy of the new proposition. The method has the advantage that the content generated by the large model can be compared with the authoritative data source, so that the accuracy of error identification is improved.
S4, automatically correcting errors in the new proposition according to the analysis result and verification feedback, and finally realizing effective error correction of the new proposition. Based on the analysis and verification of the previous steps, the system automatically proposes correction advice. For example, for the new proposition regarding global air temperature changes described above, the system may suggest a modification to "global air temperature has risen by X degrees over the past decade. "the X-degree here is calculated from the data in the domain-specific knowledge base. In addition, the modification process may include adjustments to portions of ambiguous or logically inconsistent semantics to ensure that new propositions are both accurate and clear. By the method, errors can be corrected, and quality of new propositions can be improved, so that the new propositions are more in line with actual situations.
Through the steps, the problem of 'how to effectively identify and correct errors in new propositions generated by a large model' can be solved, the whole flow starts from the acquisition of the new propositions and the context information thereof, potential errors are identified through semantic analysis, the accuracy is verified by using domain knowledge, and finally, the errors are automatically corrected, so that a complete error correction mechanism is formed.
Next, specific steps for acquiring the generated new proposition and its corresponding context information are described:
And receiving the output new proposition text through the API interface. This process involves interactions with the large model, typically by calling a predefined application programming interface API. Specifically, after a new segment of proposition text is generated, the text is transmitted to the error correction system through the API interface for further processing and analysis.
And extracting key words in the new proposition text. After receiving the new proposition text, the system needs to perform preliminary analysis on the text content to identify key information therein. This step can be accomplished through a variety of natural language processing techniques, such as part-of-speech tagging, named entity recognition, etc., to screen out words that are critical to understanding the meaning of text.
Context is built based on the key words. To more accurately understand the meaning of a new proposition and the context in which it is located, the system uses the extracted key words to construct a context. This may involve analyzing the relevance between key words and combining knowledge bases of related fields to enhance understanding of the text context.
Entities and relationships in the new proposition are identified using natural language processing techniques. Finally, the system also needs to analyze the entity information in the new proposition and its relationship with each other in depth. This step may be accomplished by means of advanced natural language processing techniques such as dependency syntactic analysis, semantic role labeling, etc., in order to more fully understand the content of the new proposition and its underlying meaning.
Next, the specific steps of this process involved in semantic analysis of the new proposition based on the pre-trained language big model are described:
Semantic vector representations are obtained by inputting new propositions into a pre-trained language big model. This process involves passing the new proposition to be analyzed as input to a pre-trained language model. The language model is typically a deep learning model trained on large amounts of text data that captures complex structural and semantic information in the language. When a new proposition is entered, the model processes it and generates one or more artificial vectors, which can be regarded as semantic representations of the new proposition.
By calculating the similarity S between the new proposition and the in-domain standard expression based on the semantic vector representation. After the semantic vector representation of the new proposition is obtained, the similarity between this representation and standard expressions known in the art needs to be calculated next. This step may be accomplished in a number of ways, such as using cosine similarity, euclidean distance, or other metrics to quantify the degree of similarity between the two vectors. The value of similarity S reflects the semantically close degree of the new proposition to the standard expression.
If the similarity S is lower than a preset threshold T, the fact that the new proposition possibly has semantic errors is judged. After the similarity S is calculated, it is compared with a threshold T set in advance. If S is less than T, the new proposition is considered to have larger semantic difference with the standard expression in the field, and the semantic error or inaccurate expression can be contained. The choice of this threshold depends on the specific application scenario and the requirements for error tolerance.
And comparing the new propositions with key difference points of standard expressions in the field. Once it is determined that the new proposition may have semantic errors, it is then necessary to further analyze key differences between the new proposition and the standard expression. This can be done by comparing the semantic vector representations of the two to find out the specific cause that results in a lower similarity. For example, places where a particular word or phrase in the new proposition does not agree with the standard expression may be identified and modification suggestions may be made accordingly to improve the accuracy of the new proposition.
Next, the specific steps of calculating the similarity S between the new proposition and the in-domain standard expression based on the semantic vector representation are described:
First, a new proposition is semantically analyzed to obtain its semantic vector representation, and normalized. This process ensures that vectors of different lengths can be compared under a uniform standard, thereby avoiding the influence of vector length differences on the similarity calculation. Normalization is typically done using an L2 norm or other normalization method.
Then, the cosine similarity C between the normalized new propositional semantic vector representation and the in-domain standard expression vector representation is calculated. Cosine similarity is a method for measuring cosine value of included angle of two non-zero vectors, and is used for evaluating direction similarity between two non-zero vectors. The cosine similarity C is obtained by calculating the product of the two vector dot products divided by the respective modulo length.
Then, judging whether the cosine similarity C is smaller than a threshold value of 0.7, and if the cosine similarity C is smaller than 0.7, considering that the new proposition has larger difference with the standard expression in terms of semantics, namely mismatching. It should be noted that, in practical applications, the range of cosine similarity is [ -1, 1], so that the threshold value 0.7 may be a pen error or a misexpression, and the normal threshold value should be a suitable value selected within this range.
Finally, the calculation mode of the similarity S is adjusted based on the mismatch condition. For example, other factors such as vocabulary overlapping degree, syntax structure similarity and the like can be introduced to comprehensively evaluate the similarity S between the new proposition and the standard expression, or the threshold value of cosine similarity can be adjusted to adapt to different application scenes and requirements.
Next, the specific steps of calculating the cosine similarity C between the normalized semantic vector representation and the intra-domain standard expression vector representation are described:
determining new propositional semantic vector representation V1 and standard expression vector representation V2
In this step, first, a semantic vector representation V1 of a new proposition after preprocessing and normalization processing and a vector representation V2 of a standard expression in the field need to be acquired. These vector representations are typically generated by pre-trained language models that capture semantic information in text. For example, the semantic features of text may be extracted using BERT, ROBERTA, or other similar deep learning models and converted into a fixed length vector form.
Calculating the dot product P of V1 and V2;
Next, the dot product P between the two vectors V1 and V2 obtained above is calculated. Dot product is a mathematical operation that measures the sum of two vectors after multiplication of elements in the same dimension. Specifically, if V1 and V2 are both n-dimensional vectors, the dot product P may be obtained by multiplying the ith element of V1 by the ith element of V2 and summing, namely:
;
Where V 1 (i) represents the i-th element of vector V 1, V 2 (i) represents the i-th element of vector V 2, n is the dimension of the vector, x represents the product of the elements, and Σ represents the sum from the 1 st to the n-th element.
Calculating the module lengths M1 and M2 of V1 and V2;
After the dot product is calculated, the modulo lengths M1 and M2 of the two vectors V1 and V2, respectively, need to be further calculated. The modulo length, or norm, of a vector is an indicator of the length of the vector, and can be obtained by summing the squares of the components of the vector and then dividing the square root. For the vector V1, the module length M1 is calculated as follows;
;
for vector V2, the modulus M2 is calculated as follows.
;
Calculating cosine similarity C;
the final step is to calculate the cosine similarity C, which is obtained by dividing the dot product P by the modulo-length product m1×m2 of the two vectors. The calculation formula of the cosine similarity C is c=p/(m1×m2). The cosine similarity has a value in the range of-1 to 1, where 1 indicates that the two vectors are identical, 0 indicates that the two vectors are orthogonal, i.e., have no correlation, and-1 indicates that the two vectors are opposite in direction. In this way we can quantify the degree of semantic similarity between the new proposition and the standard expressions within the domain.
Next, specific steps for judging the matching degree of the new proposition and the standard expression with respect to the cosine similarity will be described:
By computing cosine similarity between the new proposition and the standard expression, first, the new proposition and the standard expression are respectively converted into vector representation forms, which can be accomplished by a pre-trained semantic model. Then, cosine similarity C between the two is calculated using these vectors. Cosine similarity is a measure of the angle between two non-zero vectors, with values ranging from-1 to 1, with values closer to 1 indicating that the two vectors are more similar.
And judging the matching degree of the new proposition and the standard expression by setting a threshold value, namely, when the calculated cosine similarity C is smaller than 0.7, namely, C is smaller than 0.7, considering that the new proposition and the standard expression have larger difference, namely, are not matched. This means that the new proposition is semantically significantly different from the standard expression, possibly because the new proposition contains some erroneous or inaccurate information.
And (3) taking corresponding processing measures according to the judgment result, wherein once the large difference exists between the new proposition and the standard expression, the system can further analyze the specific problem in the new proposition and provide correction suggestions or directly give out the correct standard expression as a reference so as to help a user to understand and correct errors in the new proposition.
Next, the specific steps of the recognition process for semantic analysis to identify potential errors using the pre-trained language big model are described:
The lexical structure information of the new proposition is acquired, at this stage, the input new proposition is first received and preliminary processing is performed thereon to extract the basic vocabulary units constituting the proposition and the positions and actions thereof in the sentence. This step typically involves parsing the sentence structure using natural language processing techniques, such as by dependency or component syntactic analysis to determine the relationships between the individual words and their functional roles in the sentence.
The grammar probability score P (grammar) for the new proposition is calculated based on the language model, and then the grammar correctness of the entire new proposition is evaluated using the pre-trained language model. This process involves passing a new proposition to the language model, which computes the likelihood of the proposition as a legitimate sentence from its internally learned probability distribution. The resulting grammar probability score P (grammar) reflects the rationality of the proposition at the grammar level.
The system then checks whether the calculated P (grammar) value is below a preset threshold θ1 by determining whether the grammar probability score P (grammar) is below a first preset threshold θ1. If P (grammar) is less than θ1, it indicates that the new proposition may have an error at the grammar level. The selection of the threshold value needs to be determined according to the specific application scene and the capability of the language model, so that the error can be effectively detected and the false alarm can be avoided.
If P (grammar) < θ1, then a grammar error is likely to exist for the new proposition, and when P (grammar) is indeed below θ1, the system marks the new proposition as a sentence that may contain a grammar error. At this point, further measures may be taken to correct these errors, such as providing modification suggestions or directly applying an auto-correcting algorithm to improve the grammar structure of the sentence.
Through the steps, the method and the device can effectively utilize the strong capability of the pre-training language model to identify and correct grammar errors in new propositions, thereby improving the overall quality of texts.
Next, specific steps of calculating a grammar probability score P (grammar) of a new proposition based on the language model are described:
The new proposition is subjected to word segmentation processing to obtain a series of words, namely word sequences, which form the proposition. The process utilizes word segmentation technology in natural language processing, ensures that each vocabulary unit in a new proposition can be accurately identified, and provides a basis for subsequent probability calculation.
The probability score P (vocabulary) of the vocabulary sequence is calculated by applying a pre-trained language model. In this step, the language model evaluates the probability score of the whole sequence according to the occurrence probability of each vocabulary in the vocabulary sequence and the combination relation between the vocabulary sequence, and reflects the linguistic rationality of the vocabulary sequence.
The context correlation score P (context) is calculated by analyzing the degree of association between the new proposition and the context. In particular, techniques such as attention mechanisms can be employed to measure semantic consistency and consistency between new propositions and contexts to quantify whether it is reasonable in a particular context.
Finally, a grammar probability score P (grammar) for the new proposition is comprehensively calculated by the formula P (grammar) =p (vocabulary) ×p (context). This means that the grammar correctness of the new proposition is not only dependent on the rationality P (vocabulary) of its internal vocabulary sequence, but also on the degree P (context) to which it matches the context environment. This comprehensive evaluation approach helps to more fully determine the grammatical quality of the new proposition.
Next, specific steps of calculating a context correlation score P (context) of a new proposition are described:
This process begins by obtaining contextual information of the new proposition to be processed. This step involves identifying text segments immediately adjacent to the new proposition, which constitute the context of the new proposition. For example, in a continuous piece of text, if the new proposition is sentence "B", sentence "a" as the front and sentence "C" as the back will be extracted for subsequent analysis.
Relevance scores P (front) and P (rear) of the context information are calculated by using a pre-trained language model. Specifically, the language model evaluates the semantic continuity between sentence "a" and sentence "B" and the semantic continuity between sentence "C" and sentence "B" and gives a numerical value representing the degree of such continuity, respectively. For example, advanced language models such as BERT or GPT may be used to make such calculations.
Next, it is determined whether both the obtained correlation scores P (before) and P (after) are higher than a second preset threshold θ2. This threshold is the lowest criterion for measuring the correlation between the context and the new proposition. If both scores exceed this threshold, the new proposition is considered to have good consistency with its context, whereas the consistency is considered to be poor. For example, assuming that θ2 is set to 0.6, if P (front) =0.75 and P (rear) =0.80, the condition is satisfied.
If P (front) > θ2 and P (back) > θ2, a context correlation score P (context) for the new proposition is calculated, i.e., P (context) =p (front) +p (back) \2. This means that the relevance scores of the new proposition and the context will be averaged to arrive at a composite score reflecting the level of consistency of the new proposition in the overall context. For example, if P (front) =0.75 and P (rear) =0.80, then P (context) =0.75+0.80/2=0.775.
Next, specific steps of calculating the relevance scores P (before) and P (after) of the context information are described:
First, the contextual keywords of the new proposition are extracted. This process involves analyzing the new proposition and its context to identify words or phrases that are closely related to the new proposition. These keywords may be nouns, verbs, or other key elements that reflect contextual meaning. For example, if the new proposition is "the largest planet in the solar system is the star", then the keywords may include "solar system", "max", and "star".
Next, a probability P (keyword) that the keyword appears in the context is calculated based on the language model. This step utilizes a pre-trained language model to evaluate the likelihood that keywords will appear in a given context. For example, in the above example, the language model may give a higher probability of occurrence of the "solar system", while the probabilities of the "maximum" and "stars" are determined according to their frequency of occurrence in the context.
Then, a semantic similarity S (semantic) between the keyword and the new proposition is calculated. This step typically involves using a semantic similarity algorithm to measure the strength of the relationship between the keywords and the new proposition. For example, cosine similarity or the like can be used to quantify the degree of semantic association between keywords and new propositions.
Finally, the relevance scores P (front) and P (rear) are calculated according to the formula P (front)/P (rear) =p (keyword) ×s (semantic). This means that the probability of the keyword appearing in the context is multiplied by its semantic similarity to the new proposition to obtain a composite score representing the relevance of the keyword to the new proposition in the context. For example, if the probability of the "solar system" occurring in context is high and there is a strong semantic association with a new proposition, it will get a higher relevance score.
Next, specific steps of calculating the semantic similarity S (semantics) between the keyword and the new proposition are described:
the keywords and new propositions are respectively vectorized and represented as a vector V keyword and a vector V new propositions. This process typically involves converting keywords and new propositions in text into a digitized vector form using pre-trained language models or word embedding techniques. For example, word2Vec, gloVe, or more advanced BERT, etc. models may be used to obtain these vectors. These models can capture the meaning that the vocabulary carries in different contexts and map it into a multidimensional space such that semantically similar words are closer together in that space.
And calculating cosine similarity COS between the vector V key words and the V new proposition. Cosine similarity is a measure of the angle between two non-zero vectors used to evaluate their directional consistency. Specifically, the calculation can be performed by the formula cos=v keyword, V new proposition, x new proposition, wherein, representing a vector point multiplication operation, V keywords and V new propositions represent the modulo, i.e., the length, of the two vectors, respectively. The COS values thus obtained range from-1 to 1, the closer the values to 1 the more similar the two vectors.
If COS > the third preset threshold θ3, S (semantic) =cos. A threshold value theta 3 is introduced as a judgment standard, and only when the calculated cosine similarity COS is larger than the threshold value, the higher semantic similarity exists between the keywords and the new proposition, and the COS is assigned to S (semantic). For example, assuming θ3 is set to 0.7, keywords are considered to have a semantically strong correlation with new propositions only when COS is greater than 0.7. The setting is helpful to filter out the keyword combinations with larger semantic differences, thereby improving the effectiveness and accuracy of the new proposition error correction method.
In actual operation, when the device is in use, new propositions and their associated context information from the large model generation are first received through the input interface. This information is then passed to a language model module that performs in-depth semantic analysis of the new proposition based on the pre-trained language model to identify grammatical or logical potential errors that may exist therein. Meanwhile, the domain-specific knowledge base is taken as another key component, and contains expertise and data related to the specific domain for further verifying the accuracy of the new proposition. Once the language model module has completed the preliminary semantic analysis and marked the potential errors, this information is sent to a knowledge verification module that uses the data in the domain-specific knowledge base to confirm or negate the potential errors. If errors are found to do exist in the new proposition, the system will submit these errors to the error correction algorithm module along with the original new proposition. The error correction algorithm module automatically corrects errors in the new proposition according to the received information, the previous analysis result and the verification feedback, and generates a corrected proposition. Finally, the corrected new propositions are output, the whole process realizes the automatic flow from receiving the new propositions to outputting the propositions accurately and without errors, and the quality and the accuracy of the large model generated content are ensured. The series of steps not only improves the processing efficiency, but also ensures the specialty and reliability of the final output content.
The step S3 specifically includes:
Relevant data is extracted from a predefined domain-specific knowledge base containing facts, rules, logical relations, and domain-specific concepts.
Let the knowledge base be K, the new topic be P,
For each sub-proposition Pi in P, find from K the set of propositions { K1, K2, & gt, kn } that is closest to Pi semantics,
Calculating the similarity of corresponding propositions Kj in propositions Pi and KThe cosine similarity formula is used:,
Wherein, AndIs a semantic vector representation of the corresponding proposition;
Verifying the logic consistency of the new proposition and the known knowledge according to rules and logic in a knowledge base, evaluating the overall accuracy of the new proposition PPP, comprehensively considering the similarity and logic consistency of all sub propositions, and obtaining the confidence score C (P) of the new proposition:
,
Wherein, Is a logical consistency score.
In this embodiment, the error correction method analyzes each sub-proposition in the new proposition by using data in the domain-specific knowledge base. First, the similarity between the sub-proposition Pi and the knowledge base proposition Kj is calculated by extracting the proposition set { K1, K2, & gt, kn } closest to each sub-proposition Pi from the knowledge base K. This similarity is calculated using a cosine similarity formula, ensuring accurate quantification of the semantic relationship between propositions. Next, the method verifies the logical consistency of the new proposition with the prior knowledge in the knowledge base based on predefined rules and logic. By combining the semantic similarity and logical consistency of the sub-propositions, the method can calculate a confidence score C (P) for the new proposition, which reflects the overall accuracy of the new proposition and consistency with known information in the knowledge base.
According to the embodiment, each sub-proposition of the new proposition is compared with the propositions in the domain-specific knowledge base, so that the verification accuracy of the propositions is effectively improved. By calculating the similarity and logic consistency of the sub-propositions, the method can comprehensively evaluate the overall accuracy of the new propositions and obtain reliable confidence scores. The process ensures consistency of new propositions in logic and semantics, thereby improving the accuracy and reliability of the error correction process.
In this embodiment, different types of similarity calculation methods, such as euclidean distance, jaccard similarity coefficient, etc., may be selected to replace cosine similarity, and adjusted according to requirements in different application scenarios. In addition, the construction mode of the knowledge base K can also be changed, for example, more domain-specific corpuses are introduced or the dynamic updating capability of the knowledge base is enhanced so as to adapt to a continuously-changing knowledge system. The final confidence score C (P) may also be optimized by different logic rules or weight adjustments to further improve the accuracy of the error correction.
The step S4 specifically includes:
determining the position of the error in the proposition based on the similarity and the logical consistency result in the step S3, locating the error point of each sub-proposition Pi if Less than a preset thresholdThen consider that the sub-proposition has an error;
Error correction is carried out by utilizing the correct proposition sets { k1, k2, & gt, kn } in the domain-specific knowledge base, a replacement strategy is used for selecting a proposition Kj with highest similarity with Pi and best logic consistency to replace an error part in the Pi, and the proposition after correction is carried out The method comprises the following steps:
;
New propositions to be corrected Again through knowledge base verification, ensuring the consistency of the semantics and logic, and if the revised proposition still does not meet the conditions, further adjusting until the confidence level of the propositionReaching the set threshold value.
In this embodiment, the error correction method first identifies the error location in the new proposition from the similarity and logical consistency results obtained in the previous step. When the similarity of the sub-propositions Pi is below a preset threshold, the system recognizes that the sub-propositions may contain errors. Next, the method uses a replacement strategy to select a proposition Kj which is most similar to Pi and has the best logical consistency in the correct proposition set in the knowledge base K to replace, and generates a revised proposition P'. And then, verifying the revised proposition P 'through the knowledge base again to ensure the consistency of the proposition P' in terms of semantics and logic. If the corrected proposition still does not meet the preset confidence level requirement, further adjustment is carried out until the corrected proposition meets the condition, and the embodiment can effectively locate and correct errors in the proposition, and the effectiveness of the replacement strategy is ensured by combining analysis of two aspects of similarity and logic consistency. After verification again, the corrected proposition can ensure consistency of semantics and logic, thereby improving accuracy and efficiency of error correction. Through the process, the influence of error propositions on subsequent reasoning and decision can be greatly reduced, and the overall reliability of the knowledge base system is improved.
In this embodiment, a preset similarity threshold or a logic consistency standard may be adjusted according to different application scenarios, so as to adapt to different error correction requirements. In addition, more intelligent adjustment mechanisms can be introduced into the replacement strategy, for example, the weight of similarity calculation and logic consistency analysis is dynamically optimized through a machine learning algorithm, and the accuracy and efficiency of the correction process are further improved. If the knowledge base K has a self-learning function, the proposition set can be automatically updated in the error correction process, so that the processing capacity of the system on new propositions is improved.
The invention also provides a new proposition error correction system based on a large model, which is used for realizing the new proposition error correction method based on the large model, and comprises the following steps:
the context information acquisition module is used for acquiring new propositions generated by the large model and the corresponding context information;
the semantic analysis module is used for carrying out semantic analysis on the new proposition based on the pre-trained language model and identifying potential errors in the new proposition;
The verification module utilizes the domain-specific knowledge base to verify the accuracy of the new proposition;
and the correction module automatically corrects errors in the new propositions according to the semantic analysis result and verification feedback.
The context information acquisition module is used for receiving and processing new propositions generated by the large model and related context information thereof. The module captures sentences output by the large model through a dialogue or text generation system and extracts relevant contexts to ensure the accuracy of subsequent processing.
The semantic analysis module performs deep semantic analysis on the new propositions through a pre-trained language model. This module identifies potential errors in the new proposition, including in particular but not limited to grammatical errors, logical contradictions, or semantic inconsistencies, etc. The core of the semantic analysis module is to comprehensively analyze the semantics of the new proposition by utilizing the context understanding capability of the language model so as to ensure the integrity of the semantics in logic and grammar.
The verification module verifies the content of the new proposition by calling the domain-specific knowledge base. These knowledge bases may contain art-recognized standards, rules, or known factual information. The module is responsible for detecting whether the new proposition is consistent with the knowledge known in the art and marking potential inaccuracy.
The correction module automatically corrects errors in the new proposition based on the semantic analysis result and verification feedback. The process of correction relies on understanding the semantics of the model and combining the results of the verification in the domain knowledge base to generate a new proposition that is error corrected, consistent with the context and domain knowledge. The context information acquisition module may also employ different acquisition strategies, for example, by monitoring multiple sources (e.g., document generation systems, dialog agents) in real time to collect context information. In addition, the semantic analysis module can adopt different pre-training language models according to different application scenes so as to adapt to specific language styles or field requirements. The validation module may also incorporate a variety of knowledge bases or use external APIs to validate in the absence of knowledge base support. The rework module may also incorporate manual intervention, allowing the user to review and confirm the advice prior to automatic rework.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (7)
1. A new proposition error correction method based on a large model, the method comprising:
S1, acquiring generated new propositions and corresponding context information thereof;
s2, carrying out semantic analysis on the new propositions based on the pre-trained language big model, and identifying potential errors in the new propositions;
s3, verifying the accuracy of the new proposition by using a domain specific knowledge base;
s4, automatically correcting errors in the new propositions according to the analysis result and verification feedback;
the step S1 of obtaining the generated new proposition and the corresponding context information thereof specifically comprises the following steps:
Receiving the output new proposition text through an API interface;
Extracting key words in the new proposition text;
Building a context based on the key words;
identifying entities and relationships in the new proposition using natural language processing techniques;
the calculating the similarity S between the new proposition and the standard expression in the field based on the semantic vector representation specifically comprises the following steps:
normalizing the semantic vector representation of the new proposition;
Calculating cosine similarity C between the normalized semantic vector representation and the intra-domain standard expression vector representation;
If the cosine similarity C is smaller than 0.7, the new proposition is not matched with the standard expression, namely when C is smaller than 0.7, the new proposition is judged to have larger difference from the standard expression;
Adjusting the calculation mode of the similarity S based on the mismatch condition;
the concrete steps of calculating the cosine similarity C between the normalized semantic vector representation and the standard expression vector representation in the field include:
determining a new propositional semantic vector representation V1 and a standard expression vector representation V2;
Calculating the dot product P of V1 and V2;
Calculating the module lengths M1 and M2 of V1 and V2;
calculating cosine similarity according to the formula cosine similarity c=p/(m1×m2);
the step S3 specifically includes:
Extracting relevant data from a predefined domain-specific knowledge base containing facts, rules, logical relationships, and domain-specific concepts;
let the knowledge base be K, the new topic be P,
For each sub-proposition Pi in P, find from K the set of propositions { K1, K2, & gt, kn } that is closest to Pi semantics,
Calculating the similarity of corresponding propositions Kj in propositions Pi and KThe cosine similarity formula is used:,
Wherein, AndIs a semantic vector representation of the corresponding proposition;
Verifying the logic consistency of the new proposition and the known knowledge according to rules and logic in a knowledge base, evaluating the overall accuracy of the new proposition P, comprehensively considering the similarity and logic consistency of all sub propositions, and obtaining the confidence score C (P) of the new proposition:
,
Wherein, Is a logical consistency score.
2. The method for correcting new propositions based on large models according to claim 1, wherein said step S2 of performing semantic analysis on said new propositions based on pre-trained language large models specifically comprises:
inputting the new proposition to the pre-trained language big model to obtain a semantic vector representation;
calculating the similarity S between the new proposition and the standard expression in the field based on the semantic vector representation;
if the similarity S is lower than a preset threshold T, determining that the new proposition possibly has semantic errors;
and comparing the new propositions with key difference points of standard expressions in the field.
3. The method for correcting new propositions based on large models according to claim 1, wherein the step S2 of performing semantic analysis on the new propositions based on the pre-trained language large models comprises the following specific steps:
acquiring lexical structure information of new propositions;
Calculating a grammar probability score P (grammar) of the new proposition based on a language model;
Judging whether the grammar probability score P (grammar) is lower than a first preset threshold value theta 1 or not;
If P (grammar) < θ1, identifying that a new proposition may have a grammar error;
The specific steps of calculating the grammar probability score P (grammar) of the new proposition based on the language model comprise the following steps:
Word segmentation processing is carried out on the new proposition to obtain a vocabulary sequence;
calculating a probability score P (vocabulary) of the vocabulary sequence based on the language model;
calculating a context correlation score P (context) for the new proposition;
the grammar probability score P (grammar) is calculated according to the formula P (grammar) =p (vocabulary) ×p (context).
4. A new proposition error correction method based on big model according to claim 3, characterized in that said specific step of calculating a context correlation score P (context) of a new proposition comprises:
acquiring the context information of the new proposition;
Calculating relevance scores P (front) and P (rear) of the context information based on the language model;
judging whether the correlation scores P (front) and P (rear) are higher than a second preset threshold value theta 2 or not;
If P (front) > θ2 and P (rear) > θ2, then P (context) = (P (front) +p (rear))/2;
wherein, the specific steps of calculating the relevance scores P (front) and P (back) of the context information based on the language model comprise:
Extracting the context keywords of the new proposition;
Calculating the probability P (key words) of the key words occurring in the context based on a language model;
calculating semantic similarity S (semantics) between the keywords and the new propositions;
the correlation scores P (front) and P (rear) are calculated according to the formula P (front)/P (rear) =p (keyword) ×s (semantic).
5. The new proposition error correction method based on the big model according to claim 4, wherein the specific step of calculating the semantic similarity S (semantics) between the keyword and the new proposition comprises:
the keyword and the new proposition are respectively and vectorized to be represented as a vector V keyword and a vector V new proposition;
calculating cosine similarity COS between the vector V key words and the V new proposition;
If COS > the third preset threshold θ3, S (semantic) =cos.
6. The method for error correction of new propositions based on large models according to claim 1, wherein said step S4 specifically comprises:
determining the position of the error in the proposition based on the similarity and the logical consistency result in the step S3, locating the error point of each sub-proposition Pi if Less than a preset thresholdThen consider that the sub-proposition has an error;
Error correction is carried out by utilizing the correct proposition sets { k1, k2, & gt, kn } in the domain-specific knowledge base, a replacement strategy is used for selecting a proposition Kj with highest similarity with Pi and best logic consistency to replace an error part in the Pi, and the proposition after correction is carried out The method comprises the following steps:
;
New propositions to be corrected Again through knowledge base verification, ensuring the consistency of the semantics and logic, and if the revised proposition still does not meet the conditions, further adjusting until the confidence level of the propositionReaching the set threshold value.
7. A large model based new proposition error correction system for implementing the steps of the large model based new proposition error correction method according to any of claims 1 to 6, characterized in that the system comprises:
The context information acquisition module is used for acquiring the generated new propositions and the corresponding context information thereof;
the semantic analysis module is used for carrying out semantic analysis on the new proposition based on the pre-trained language big model and identifying potential errors in the new proposition;
The verification module utilizes the domain-specific knowledge base to verify the accuracy of the new proposition;
and the correction module automatically corrects errors in the new propositions according to the semantic analysis result and verification feedback.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411272928.5A CN118798201B (en) | 2024-09-12 | 2024-09-12 | New proposition error correction method and system based on large model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411272928.5A CN118798201B (en) | 2024-09-12 | 2024-09-12 | New proposition error correction method and system based on large model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118798201A CN118798201A (en) | 2024-10-18 |
| CN118798201B true CN118798201B (en) | 2025-01-14 |
Family
ID=93031373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411272928.5A Active CN118798201B (en) | 2024-09-12 | 2024-09-12 | New proposition error correction method and system based on large model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118798201B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119514553B (en) * | 2024-11-05 | 2025-08-26 | 北京光年无限科技有限公司 | An intelligent parsing method for junior high school English reading comprehension test questions based on a large model |
| CN119476246A (en) * | 2025-01-15 | 2025-02-18 | 可之(宁波)人工智能科技有限公司 | Proposition method, system, medium and product combining big model and reasoning engine |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139387A (en) * | 2020-01-17 | 2021-07-20 | 华为技术有限公司 | Semantic error correction method, electronic device and storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11593557B2 (en) * | 2020-06-22 | 2023-02-28 | Crimson AI LLP | Domain-specific grammar correction system, server and method for academic text |
| CN116702760A (en) * | 2023-06-01 | 2023-09-05 | 中国石油大学(华东) | A Geographic Named Entity Error Correction Method Based on Pre-training Deep Learning |
| CN117454906B (en) * | 2023-12-22 | 2024-05-24 | 创云融达信息技术(天津)股份有限公司 | Text proofreading method and system based on natural language processing and machine learning |
-
2024
- 2024-09-12 CN CN202411272928.5A patent/CN118798201B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139387A (en) * | 2020-01-17 | 2021-07-20 | 华为技术有限公司 | Semantic error correction method, electronic device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118798201A (en) | 2024-10-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN118798201B (en) | New proposition error correction method and system based on large model | |
| CN119808917B (en) | A knowledge graph construction method based on fine-tuning large language model | |
| CN117454906B (en) | Text proofreading method and system based on natural language processing and machine learning | |
| CN110008323B (en) | A method for problem equivalence discrimination of semi-supervised learning combined with ensemble learning | |
| CN113743099A (en) | Self-attention mechanism-based term extraction system, method, medium and terminal | |
| CN113705207A (en) | Grammar error recognition method and device | |
| CN118152428A (en) | A method and device for predicting and enhancing query instructions of power customer service system | |
| CN119514553A (en) | An intelligent parsing method for junior high school English reading comprehension test questions based on a large model | |
| CN120104765A (en) | A large language model hallucination detection method and system based on knowledge graph structure | |
| CN119337850A (en) | Method, device, electronic device and storage medium for constructing bid document data set | |
| US12253988B1 (en) | Text analysis and verification methods and systems | |
| CN114036955B (en) | Detection method for headword event argument of central word | |
| CN113158675B (en) | Entity extraction method, device, equipment and medium based on artificial intelligence | |
| CN114896966A (en) | Method, system, equipment and medium for positioning grammar error of Chinese text | |
| CN120929601B (en) | A Deep Learning-Based Intelligent Classification Method and System for Natural Language Text Data | |
| CN113987141B (en) | Question-answering system answer reliability instant checking method based on recursion inquiry | |
| CN113822052A (en) | A text error detection method, device, electronic device and storage medium | |
| CN120407775A (en) | A method and system for automatic annotation of text data integrating pre-trained NLP models | |
| CN119577622A (en) | A Chinese electric power text error detection method and system based on ALBert model | |
| KR20220021836A (en) | Context sensitive spelling error correction system or method using Autoregressive language model | |
| CN117875706A (en) | An AI-based digital management method for grading process | |
| CN117556789A (en) | Student review generation method based on multi-level semantic mining | |
| CN114239555A (en) | Training method of keyword extraction model and related device | |
| Zhou et al. | RNN-based sequence-preserved attention for dependency parsing | |
| CN119670732B (en) | A knowledge-enhanced text correction method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |