Text simplification method and system based on retrieval enhancement generation

Abstract

The invention provides a text simplification method and system based on retrieval enhancement generation, and relates to the technical field of text simplification. According to the method, the to-be-simplified sentence is obtained, and the complex words in the to-be-simplified sentence are recognized; retrieving a plurality of possible paraphrases of the complex words, respectively replacing the complex words in the to-be-simplified sentence with the possible paraphrases to form a plurality of paraphrasing sentences, and selecting the paraphrasing sentence with the highest similarity with the to-be-simplified sentence from the plurality of paraphrasing sentences as an optimal paraphrasing sentence; filling the sentence to be simplified and the optimal paraphrasing sentence into a predefined prompt template to obtain an input sentence converted by the prompt template; and taking the input sentence as input data of the large language model, and outputting the simplified sentence. According to the method, the complex text paraphrases related to the semantics of the content to be simplified are retrieved, so that the text simplification accuracy can be improved, and the relevance of the text and the integrity of information can be ensured to be kept.

Description

Technical Field

[0001] The present invention relates to the technical field of text simplification, and in particular to a text simplification method and system based on retrieval enhanced generation. Background Art

[0002] Text simplification refers to the process of reducing the complexity of text while retaining the original information and meaning of the sentence, including the optimization of complex sentence structure, the removal of redundant information, and the simplification of complex words. Generally, large language models are not trained for text simplification tasks. Retraining large language models requires a lot of manpower and material resources, which makes large language models often have poor simplification effects for text simplification tasks and inaccurate simplified content because they do not have enough simplified information.

[0003] Existing text simplification methods mainly rely on large language models such as Llama. These models are usually fine-tuned to suit specific application scenarios. In this way, the model can understand and reorganize the structure of complex text to generate a simplified version that is semantically coherent and easier to understand.

[0004] However, although large language models can acquire a certain text simplification capability after being trained with training data, this capability still has limitations. This is mainly reflected in the fact that when the model processes complex sentences, especially texts with specific industry terms or highly specialized content, it often fails to achieve the ideal simplification effect due to the failure to fully understand the deep meaning and contextual relationships of these complex words. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the deficiencies in the prior art, the present invention provides a text simplification method and system based on retrieval enhanced generation, which solves the technical problem that the existing text simplification methods cannot fully understand the deep meaning and contextual relationship of complex words in the text when processing texts with specific industry terms or highly specialized content.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] In a first aspect, the present invention provides a text simplification method based on retrieval enhancement generation, comprising:

[0010] Obtaining a sentence to be simplified, and identifying complex words in the sentence to be simplified;

[0011] Retrieve several possible interpretations of complex words, replace the complex words in the sentence to be simplified with the possible interpretations, form several interpretation sentences, and select the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence;

[0012] Fill the sentence to be simplified and the optimal interpretation sentence into the predefined prompt template to obtain the input sentence converted by the prompt template;

[0013] The input sentence is used as input data for a large language model and the simplified sentence is output.

[0014] Preferably, the step of identifying complex words in the sentence to be simplified includes:

[0015] The sentences to be simplified are processed through the word frequency database to identify the complex words in the sentences to be simplified.

[0016] Preferably, when the sentence to be simplified is in English, before performing the retrieval of several possible interpretations of a complex word, the text simplification method further comprises: restoring the basic form of the complex word by a natural language processing toolkit.

[0017] Preferably, the retrieving several possible interpretations of a complex word includes: searching for all possible interpretations of the complex word through a dictionary.

[0018] Preferably, the step of replacing the complex words in the sentence to be simplified with possible interpretations to form a plurality of interpretation sentences, and selecting the interpretation sentence with the highest similarity to the sentence to be simplified from the plurality of interpretation sentences as the optimal interpretation sentence includes:

[0019] For a sentence p to be simplified containing complex words, convert it into a vector p in the embedding space embed ;

[0020] p embed =Embed(p)

[0021] Replace the complex words and sentences in the sentence to be simplified with various possible interpretations, and calculate the embedding vector c of the newly generated interpretation sentence after each replacement i and the sentence to be simplified is embedded in p embed The similarity between

[0022] Select the paraphrase sentence with the highest similarity to the sentence to be simplified p as the optimal paraphrase sentence R.

[0023] Preferably, the selecting the paraphrase sentence with the highest similarity to the sentence to be simplified from the plurality of paraphrase sentences as the optimal paraphrase sentence comprises: calculating the similarity between the paraphrase sentence and the sentence to be simplified based on a pre-constructed similarity measurement model;

[0024] The pre-built similarity measurement model adopts a contrastive learning framework and uses supervised contrastive learning to explicitly define which examples should be attracted or repelled. During the training process, sentences with high readability are regarded as positive samples, and sentences with low readability are regarded as negative samples. In terms of form, a quadruple is constructed. in and h i are two different highly readable sentences, and l i are two different low-readability sentences; two optimization objectives are defined when training the similarity measurement model:

[0025] Optimization goal 1: Bring sentences with low readability closer to sentences with high readability:

[0026]

[0027] Where sim(·) represents the cosine similarity between sentence embeddings, α represents the weight of negative samples, τ is the temperature parameter, i represents the index of a quadruple in the current batch, j represents the index of other quadruple different from i, and N represents the total number of quadruple;

[0028] Optimization goal 2: Push low-readability sentences away from each other:

[0029]

[0030] By optimizing the loss function L s , the embedding learned by the similarity metric model can bring sentences with similar readability closer together, while pushing non-adjacent sentences with different readability farther apart. The calculation formula is as follows:

[0031]

[0032] In a second aspect, the present invention provides a text simplification system based on retrieval enhancement generation, comprising:

[0033] A complex word recognition module is used to obtain a sentence to be simplified and recognize complex words in the sentence to be simplified;

[0034] The optimal interpretation sentence selection module is used to retrieve several possible interpretations of complex words, replace the complex words in the sentence to be simplified with various possible interpretations, form several interpretation sentences, and select the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence;

[0035] A template filling module is used to fill the sentence to be simplified and the optimal interpretation sentence into a predefined prompt template to obtain an input sentence converted by the prompt template;

[0036] The output module is used to take the input sentence as input data of the large language model and output the simplified sentence.

[0037] Preferably, when the sentence to be simplified is in English, before performing the retrieval of several possible interpretations of the complex word, the optimal interpretation sentence selection module is further used to: restore the basic form of the complex word through a natural language processing toolkit.

[0038] In a third aspect, the present invention provides a storage medium storing a computer program for text simplification based on retrieval enhancement generation, wherein the computer program enables a computer to execute the text simplification method based on retrieval enhancement generation as described above.

[0039] In a fourth aspect, the present invention provides an electronic device, comprising:

[0040] one or more processors;

[0041] Memory; and

[0042] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including a method for executing the text simplification method based on retrieval enhancement generation as described above.

[0043] (III) Beneficial effects

[0044] The present invention provides a text simplification method and system based on retrieval enhancement generation. Compared with the prior art, it has the following beneficial effects:

[0045] The present invention provides a text simplification method based on retrieval enhancement generation, which obtains a sentence to be simplified and identifies complex words in the sentence to be simplified; retrieves several possible interpretations of the complex words, replaces the complex words in the sentence to be simplified with each possible interpretation, forms several interpretation sentences, selects the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence; fills the sentence to be simplified and the optimal interpretation sentence into a predefined prompt template to obtain an input sentence converted by the prompt template; uses the input sentence as input data of a large language model, and outputs the simplified sentence. The present invention can not only improve the accuracy of text simplification by retrieving complex text interpretations related to the semantics of the content to be simplified, but also ensure that the relevance of the text and the integrity of the information are maintained. Especially when processing texts containing professional terms or with strong technicality, this method combining generation and retrieval is significantly superior to the traditional large language model simplification method, because it can accurately process and interpret professional content, thereby generating more accurate and easy-to-understand simplified texts. In addition, this method also has high adaptability and can be flexibly adjusted according to different fields and content requirements, further improving its practical value in a variety of application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A block diagram of a text simplification method based on retrieval enhancement generation according to an embodiment of the present invention;

[0048] Figure 2 Schematic diagram of the specific process of the text simplification method based on retrieval enhancement generation. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The embodiments of the present application provide a text simplification method and system based on retrieval-enhanced generation, which solves the technical problem that existing text simplification methods cannot fully understand the deep meaning and contextual relationships of complex words in the text when processing texts with specific industry terms or highly specialized content, thereby achieving accurate processing and interpretation of professional content, thereby generating more accurate and easy-to-understand simplified text.

[0051] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0052] In the current technical context, text simplification mainly relies on large language models such as Llama. These models are usually fine-tuned to adapt to specific application scenarios. In this way, the model is able to understand and reorganize the structure of complex texts to generate a simplified version that is semantically coherent and easier to understand. Although large language models can acquire certain text simplification capabilities through training data, this capability still has limitations. Models often fail to achieve ideal simplification effects when processing complex sentences, especially texts with specific industry terms or highly specialized content. This is because the model may fail to fully understand the deep meaning and contextual relationships of these complex words.

[0053] In order to solve the above problems, the embodiment of the present invention proposes a text simplification method and system based on retrieval enhancement generation, aiming to overcome the limitations of the existing technology in terms of difficulty in obtaining data sets and processing professional content. This method effectively improves the application efficiency and accuracy of text simplification in professional fields by combining advanced language models and retrieval technology.

[0054] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0055] The embodiment of the present invention provides a text simplification method based on retrieval enhancement generation, such as Figure 1 As shown, including:

[0056] S1. Obtain a sentence to be simplified and identify complex words in the sentence to be simplified;

[0057] S2, searching for several possible interpretations of the complex word, replacing the complex word in the sentence to be simplified with each possible interpretation, forming several interpretation sentences, and selecting the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence;

[0058] S3, filling the sentence to be simplified and the optimal interpretation sentence into a predefined prompt template to obtain an input sentence converted by the prompt template;

[0059] S4. Use the input sentence as input data for a large language model and output a simplified sentence.

[0060] The core of the embodiment of the present invention is to adopt a new text simplification strategy, which combines a language model and a dynamic content retrieval system. By retrieving complex text interpretations that are semantically related to the content to be simplified, not only can the accuracy of text simplification be improved, but also the relevance of the text and the integrity of the information can be maintained. Especially when dealing with texts containing professional terms or highly technical texts, this combined generation and retrieval method is significantly superior to traditional large-scale language model simplification methods because it can accurately process and interpret professional content, thereby generating more accurate and easy-to-understand simplified texts. In addition, this method also has high adaptability and can be flexibly adjusted according to different fields and content requirements, further improving its practical value in a variety of application scenarios.

[0061] The following is combined with Figure 2 The specific process diagram of the text simplification method based on retrieval enhancement generation shown in the figure describes the embodiment of the present invention in detail:

[0062] In step S1, a sentence to be simplified is obtained, and complex words in the sentence to be simplified are identified. The specific implementation process is as follows:

[0063] The sentence to be simplified is processed through the word frequency database to identify the complex words in the sentence to be simplified. The word frequency database can be selected from wordfreq database, MRC Psycholinguistic Database and Dataset Kuperman 2012AoA, etc.

[0064] Take the wordfreq library as an example. The library uses Zipf to measure the frequency of vocabulary. Zipf means that the frequency of a word is the base 10 logarithm of the number of times the word appears in one billion words. For example, a word with a Zipf value of 6 means that it appears approximately once in every thousand words. This article sets the Zipf value to 3, which means that words that appear once in every million words are considered complex words.

[0065] In step S2, several possible interpretations of complex words are retrieved, and the complex words in the sentence to be simplified are replaced with various possible interpretations to form several interpretation sentences, and the interpretation sentence with the highest similarity to the sentence to be simplified is selected from the several interpretation sentences as the optimal interpretation sentence. The specific implementation process is as follows:

[0066] S201. Search for possible interpretations of complex words, including:

[0067] Search for all possible meanings of complex words through dictionaries. You can choose any dictionary with a database on the Internet, such as Merriam-Webster's Primary Dictionary (MWED), Xinhua Dictionary, etc.

[0068] It should be noted that when the sentence to be simplified is in English, before searching for all possible meanings of a complex word in a dictionary, it is also necessary to restore the basic form of the complex word through a natural language processing toolkit, that is, convert it into a prototype in the dictionary. For example, restore the past tense of a verb to the present tense, and restore the plural form of a noun to the singular. Natural language processing toolkits can be selected from nltk libraries, SpaCy libraries, TextBlob libraries, and Gensim libraries.

[0069] S202, replacing the complex words in the sentence to be simplified with possible interpretations to form a number of interpretation sentences, and selecting the interpretation sentence with the highest similarity to the sentence to be simplified from the number of interpretation sentences as the optimal interpretation sentence, specifically including:

[0070] The core idea of ​​this part is to replace the complex words in the sentence with the previously obtained interpretation, find the replaced text that is most semantically consistent with the sentence to be simplified, and thus determine the optimal interpretation sentence. To achieve this goal, for the sentence to be simplified p containing complex words, first convert it into a vector p in the embedding space embed .

[0071] p embed =Embed(p)

[0072] The complex words and sentences in the sentence to be simplified are replaced with the interpretations retrieved in S201. After each replacement, the embedding vector c of the newly generated interpretation sentence is calculated. i and the sentence to be simplified is embedded in p embed In this embodiment of the present invention, the embedding vector c of the paraphrase sentence is calculated by using a pre-built similarity measurement model. i and the sentence to be simplified is embedded in p embed The similarity between .

[0073] The similarity metric model adopts a contrastive learning framework and uses supervised contrastive learning to explicitly define which examples should be attracted or repelled. During the training process, sentences with high readability are regarded as positive samples, and sentences with low readability are regarded as negative samples. Formally, the four-tuple is constructed in and h i are two different highly readable sentences, and l i are two different low-readability sentences. Two optimization objectives are defined when training the similarity measurement model:

[0074] 1. Bring the less readable sentences closer to the more readable sentences (i.e. make the less readable sentences closer to the more readable sentences):

[0075]

[0076] Among them, sim(·) represents the calculation of the cosine similarity between sentence embeddings, α represents the weight of negative samples, and τ is the temperature parameter. By adjusting the temperature parameter, the sensitivity of the similarity measurement model to the similarity measurement can be controlled, so that the model can adjust the accuracy and breadth of the output according to the specific situation when outputting similarity. i represents the index of a quadruple in the current batch, j represents the index of other quadruple different from i, and N represents the total number of quadruple.

[0077] 2: Push low-readable sentences away from each other:

[0078]

[0079] By optimizing the loss function L s ,The embeddings learned by the similarity metric model are able to bring sentences with similar readability closer together, while pushing non-adjacent sentences with different readability farther away.

[0080]

[0081] Finally, the paraphrase that is most semantically aligned with sentence p, that is, the paraphrase with the highest similarity, is selected as the optimal paraphrase R.

[0082] Traditional methods usually calculate sentence similarity based on pre-trained general language models, and are not specifically optimized for sentence readability. The method of the embodiment of the present invention specifically optimizes readability by defining clear quadruplets and corresponding loss functions, which helps the generated embeddings to more accurately reflect the need for sentence simplification. Specifically, the design of "pulling in" highly readable content and "pushing away" low-readable content ensures the core position of readability in similarity calculation.

[0083] Meanwhile, although traditional similarity indicators (such as cosine similarity) can reflect the similarity between sentences, they often do not clearly distinguish between complex and simple logical concepts that need to be brought closer. The similarity measurement model of the embodiment of the present invention improves the discrimination ability of the model by considering both the actions of "bringing closer" and "pushing away", so that it can make a more accurate distinction between highly complex sentences and sentences of similar complexity.

[0084] In step S3, the sentence to be simplified and the optimal interpretation sentence are filled into the predefined prompt template to obtain the input sentence converted by the prompt template. Specifically, it includes:

[0085] In this paper, the optimal paraphrase R retrieved in the previous step and the sentence to be simplified p are filled into the predefined prompt template T (i.e., large language model input prompt, including requirement description, annotations, etc.) to obtain the input sentence q. The prompt template T not only incorporates the detailed description of the specific task, but also cleverly integrates p and R, enhancing the understanding and response capabilities of the language model through carefully designed context.

[0086] q=T(p,R)

[0087] The embodiment of the present invention provides a text simplification system based on retrieval enhancement generation, comprising:

[0088] A complex word recognition module is used to obtain a sentence to be simplified and recognize complex words in the sentence to be simplified;

[0089] The optimal interpretation sentence selection module is used to retrieve several possible interpretations of complex words, replace the complex words in the sentence to be simplified with various possible interpretations, form several interpretation sentences, and select the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence;

[0090] A template filling module is used to fill the sentence to be simplified and the optimal interpretation sentence into a predefined prompt template to obtain an input sentence converted by the prompt template;

[0091] The output module is used to take the input sentence as input data of the large language model and output the simplified sentence.

[0092] It can be understood that the text simplification system based on retrieval enhancement generation provided in the embodiment of the present invention corresponds to the above-mentioned text simplification method based on retrieval enhancement generation, and the explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the text simplification method based on retrieval enhancement generation, which will not be repeated here.

[0093] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program for text simplification based on retrieval enhancement generation, wherein the computer program enables a computer to execute the text simplification method based on retrieval enhancement generation as described above.

[0094] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the text simplification method based on retrieval enhancement generation as described above.

[0095] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0096] 1. The core of the embodiment of the present invention is to adopt a new text simplification strategy, which combines a language model and a dynamic content retrieval system. By retrieving complex text interpretations that are semantically related to the content to be simplified, not only can the accuracy of text simplification be improved, but also the relevance of the text and the integrity of the information can be maintained. Especially when dealing with texts containing professional terms or highly technical texts, this combined generation and retrieval method is significantly superior to traditional large-scale language model simplification methods because it can accurately process and interpret professional content, thereby generating more accurate and easy-to-understand simplified texts. In addition, this method is also highly adaptable and can be flexibly adjusted according to different fields and content requirements, further improving its practical value in a variety of application scenarios.

[0097] 2. The embodiment of the present invention specifically optimizes readability by defining clear quadruplets and corresponding loss functions, which helps the generated embedding to more accurately reflect the need for sentence simplification. At the same time, by considering the two actions of "pulling closer" and "pushing away", the model's discrimination ability is improved, enabling it to make a more accurate distinction between high-complexity sentences and sentences of similar complexity.

[0098] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0099] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A text simplification method based on retrieval enhanced generation, characterized in that: include: Obtaining a sentence to be simplified, and identifying complex words in the sentence to be simplified; Retrieve several possible interpretations of complex words, replace the complex words in the sentence to be simplified with the possible interpretations, form several interpretation sentences, and select the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence; Fill the sentence to be simplified and the optimal interpretation sentence into the predefined prompt template to obtain the input sentence converted by the prompt template; The input sentence is used as input data for a large language model and the simplified sentence is output.

2. The text simplification method based on retrieval enhancement generation as claimed in claim 1, characterized in that: The step of identifying complex words in the sentence to be simplified includes: The sentences to be simplified are processed through the word frequency database to identify the complex words in the sentences to be simplified.

3. The text simplification method based on retrieval enhancement generation as claimed in claim 1, characterized in that: When the sentence to be simplified is in English, before searching for several possible interpretations of a complex word, the text simplification method further includes: restoring the basic form of the complex word by a natural language processing toolkit.

4. The text simplification method based on retrieval enhancement generation as claimed in claim 1, characterized in that: The retrieving several possible interpretations of the complex word includes: searching for all possible interpretations of the complex word through a dictionary.

5. The text simplification method based on retrieval enhancement generation as claimed in claim 1, characterized in that: The step of replacing the complex words in the sentence to be simplified with possible interpretations to form a plurality of interpretation sentences, and selecting the interpretation sentence with the highest similarity to the sentence to be simplified as the optimal interpretation sentence from the plurality of interpretation sentences includes: For a sentence p to be simplified containing complex words, convert it into a vector p in the embedding space embed ; p embed =Embed(p) Replace the complex words and sentences in the sentence to be simplified with various possible interpretations, and calculate the embedding vector c of the newly generated interpretation sentence after each replacement i and the sentence to be simplified is embedded in p embed The similarity between Select the paraphrase sentence with the highest similarity to the sentence to be simplified p as the optimal paraphrase sentence R.

6. The text simplification method based on retrieval enhancement generation according to any one of claims 1 to 5, characterized in that: The step of selecting the paraphrase sentence with the highest similarity to the sentence to be simplified from the plurality of paraphrase sentences as the optimal paraphrase sentence includes: calculating the similarity between the paraphrase sentence and the sentence to be simplified based on a pre-built similarity measurement model; The pre-built similarity metric model adopts a contrastive learning framework and uses supervised contrastive learning to explicitly define which examples should be attracted or repelled. During the training process, the four-tuple q i : in and h i are two different highly readable sentences, and l i are two different low-readability sentences; two optimization objectives are defined when training the similarity measurement model: Optimization goal 1: Bring sentences with low readability closer to sentences with high readability: Where sim(·) represents the cosine similarity between sentence embeddings, α represents the weight of negative samples, τ is the temperature parameter, i represents the index of a quadruple in the current batch, j represents the index of other quadruple different from i, and N represents the total number of quadruple; Optimization goal 2: Push low-readability sentences away from each other: By optimizing the loss function L s , the embedding learned by the similarity metric model can bring sentences with similar readability closer together, while pushing non-adjacent sentences with different readability farther apart. The calculation formula is as follows:

7. A text simplification system based on retrieval-enhanced generation, characterized in that: include: A complex word recognition module is used to obtain a sentence to be simplified and recognize complex words in the sentence to be simplified; The optimal interpretation sentence selection module is used to retrieve several possible interpretations of complex words, replace the complex words in the sentence to be simplified with various possible interpretations, form several interpretation sentences, and select the interpretation sentence with the highest similarity to the sentence to be simplified from the several interpretation sentences as the optimal interpretation sentence; A template filling module is used to fill the sentence to be simplified and the optimal interpretation sentence into a predefined prompt template to obtain an input sentence converted by the prompt template; The output module is used to take the input sentence as input data of the large language model and output the simplified sentence.

8. The text simplification system based on retrieval enhancement generation as claimed in claim 7, characterized in that: When the sentence to be simplified is in English, before searching for several possible interpretations of a complex word, the optimal interpretation sentence selection module is further used to restore the basic form of the complex word through a natural language processing toolkit.

9. A storage medium, characterized in that: It stores a computer program for text simplification based on retrieval enhancement generation, wherein the computer program enables the computer to execute the text simplification method based on retrieval enhancement generation as described in any one of claims 1 to 6.

10. An electronic device, characterized in that: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including a program for executing the text simplification method based on retrieval enhancement generation as described in any one of claims 1 to 6.

Images