CN117093729A - Retrieval method, system and retrieval terminal based on medical scientific research information - Google Patents

Abstract

The application provides a retrieval method, a retrieval system and a retrieval terminal based on medical scientific research information, which belong to the technical field of medical scientific research information processing and acquire retrieval information input by a user; analyzing the search information based on a natural language processing mode of the large model, analyzing sentence structures, word meaning relations and context information, and extracting keyword information; configuring the extracted keyword information into a retrieval expression; matching the medical scientific literature in the medical scientific literature database based on the retrieval expression; and displaying the matched medical scientific literature on a user interface. According to the medical scientific research-oriented retrieval method based on the large model, the intention of the user can be more accurately understood based on the natural language processing technology, and the relevance of the retrieval result is improved. And the search document can be displayed for the user, so that better user experience is provided.

Description

A retrieval method, system and retrieval terminal based on medical scientific research information

Technical field

The invention belongs to the technical field of medical scientific research information retrieval, and specifically relates to a retrieval method, system and retrieval terminal based on medical scientific research information.

Background technique

In the field of medical research, researchers need to obtain relevant information from a large amount of literature and data to support scientific research work. Traditional search methods usually require users to edit search expressions, but such methods may be difficult for non-experts, and editing expressions requires a lot of time.

The existing implementation solution most similar to the present invention is a keyword-based retrieval method. This method requires users to edit search expressions and search documents and data through keyword matching. However, this method has the following disadvantages: It is difficult for users to edit expressions: non-professionals may not be familiar with domain-related terminology and expressions, making it difficult to edit expressions; It is time-consuming: it takes a lot of time to edit complex retrieval expressions , reducing the user’s retrieval efficiency.

Disadvantages of the existing technology: Difficulty for users to edit expressions: Non-professionals may not be familiar with domain-related terminology and expressions, resulting in difficulty in editing expressions. Moreover, retrieval of medical scientific research literature takes a long time. It is necessary to edit professional vocabulary and then combine professional vocabulary before searching. If the professional vocabulary cannot form an effective search formula, it will not be possible to match the literature you want to query, which will affect scientific researchers. Experience using the system.

Contents of the invention

The invention provides a retrieval method based on medical scientific research information, which can solve the traditional problems of difficulty in editing retrieval expressions and high time consumption. The purpose of the present invention is to improve the user's retrieval efficiency and reduce the time for the user to edit expressions.

Methods include:

Step 1. Obtain the search information input by the user;

Step 2: Analyze the retrieval information based on the natural language processing method of the large model, analyze the sentence structure, word meaning relationship and contextual information, and extract the keyword information;

Among them, the large model includes: multiple identical encoder layers, each encoder layer contains a self-attention mechanism and a feed-forward neural network;

The mathematical formula of the self-attention mechanism is as follows:

Attention(Q1, K1, V1) = softmax(Q1×K1^T 1/ sqrt(d_k1)) * V1;

Among them, Q1, K1 and V1 represent the input matrix of query, key and value respectively, and d_k1 represents the dimension of the attention mechanism;

The mathematical formula of feedforward neural network is as follows:

FFN(x1) = max(0, xW_1 + b_1)W_2 + b_2;

Among them, x1 represents the input vector, W_1, b_1, W_2 and b_2 represent the parameters of the model;

The large model also includes: multiple identical decoder layers, each of which contains a self-attention mechanism and an encoder-decoder attention mechanism;

In the large model structure, the connection between the encoder layer and the decoder layer uses residual connection and layer normalization processing;

Step 3: Obtain the logical operators set by the user, combine the extracted keyword information based on the logical operators set by the user, and configure it into a search expression;

Step 4: Match the medical research literature in the medical research literature database based on the search expression;

Step 5: Display the matched medical research documents on the user interface.

It should be further noted that the method of parsing the retrieved information using natural language processing based on large models in step 2 also includes:

Segment the input natural language text into words;

Determine the part of speech of each word and extract keyword information.

It should be further noted that the method of parsing the retrieved information using natural language processing based on large models in step 2 also includes:

Analyze the sentence structure in the retrieved information, determine the dependencies between words, and extract keyword information.

It should be further noted that the method of parsing the retrieval information based on the natural language processing method of the large model in step 2 also includes: performing lexical analysis on the sentences in the retrieval information input by the user and dividing the sentences into words.

It should be further noted that lexical analysis of sentences includes: determining the dependency relationship between words in the sentence, and using dependency syntax analysis or phrase structure syntax analysis to achieve this.

It should be further noted that logical operators include: AND logic, OR logic and non-logic.

It should be further noted that step four also includes: storing multiple medical scientific research documents in the medical scientific research literature database;

Each medical research document is configured with keyword tags;

Match the keyword tags in the medical research literature database based on the search expression;

The medical scientific research documents corresponding to the matched keyword tags are displayed on the user interface.

The invention also provides a retrieval system based on medical scientific research information. The system includes: an information input module, an information analysis module, an expression configuration module, a document matching module and a document display module;

The information input module is used to obtain the retrieval information input by the user;

The information parsing module is used to parse the retrieved information in combination with the natural language processing method of the large model, parse out the sentence structure, word meaning relationships and contextual information, and extract keyword information;

Among them, the large model includes: multiple identical encoder layers, each encoder layer contains a self-attention mechanism and a feed-forward neural network;

The mathematical formula of the self-attention mechanism is as follows:

Attention(Q1, K1, V1) = softmax(Q1×K1^T 1/ sqrt(d_k1)) * V1;

Among them, Q1, K1 and V1 represent the input matrix of query, key and value respectively, and d_k1 represents the dimension of the attention mechanism;

The mathematical formula of feedforward neural network is as follows:

FFN(x1) = max(0, xW_1 + b_1)W_2 + b_2;

Among them, x1 represents the input vector, W_1, b_1, W_2 and b_2 represent the parameters of the model;

The large model also includes: multiple identical decoder layers, each of which contains a self-attention mechanism and an encoder-decoder attention mechanism;

In the large model structure, the connection between the encoder layer and the decoder layer uses residual connection and layer normalization processing; the expression configuration module is used to obtain the user settings in the logical operator, and based on the user set logic The operator combines the extracted keyword information and configures it into a search expression;

The literature matching module is used to match the medical research literature in the medical research literature database based on the search expression;

The document display module is used to display the search process information and display the matched medical scientific research documents.

The present invention also provides a retrieval terminal, including a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the medical scientific research information-based The steps of the search method.

It can be seen from the above technical solutions that the present invention has the following advantages:

In the retrieval method based on medical scientific research information involved in the present invention, the user inputs retrieval requirements through natural language expression, and the interface passes the user input to the large model for natural language understanding. The large model converts user input into retrieval expressions and passes them to the retrieval system. The retrieval system retrieves relevant medical research documents and data from the database according to the search expression, and returns the results to the user interface to display to the user. In this way, users do not need to edit complex search expressions and can search through natural language expressions, reducing users' learning costs and editing time. Moreover, the large model involved in the present invention has natural language understanding capabilities, can more accurately understand the user's intention, and improve the relevance of the search results. The user-friendly interface provides a better user experience, making retrieval easy even for non-experts.

Description of the drawings

In order to explain the technical solution of the present invention more clearly, the drawings required for the description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, As far as workers are concerned, other drawings can also be obtained based on these drawings without exerting creative work.

Figure 1 is a flow chart of the retrieval method based on medical scientific research information;

Figure 2 is a schematic diagram of a retrieval system based on medical scientific research information.

Detailed ways

The retrieval method based on medical scientific research information provided by the present invention is mainly a retrieval method for the field of medical scientific research. The method of the present invention is provided for researchers to search medical scientific research documents. Among them, in order to facilitate scientific researchers to search medical scientific research documents, the present invention can obtain and process related data based on artificial intelligence technology. Among them, methods can include technologies such as dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, etc. Retrieval method software technologies based on medical scientific research information mainly include computer perspective technology, speech processing technology, natural language processing technology, and machine learning/deep learning. Of course, for deep learning, it usually includes artificial neural network, belief network, reinforcement learning, transfer learning, inductive learning, teaching learning and other technologies. This can quickly match the medical research documents that researchers want. Specifically, it can match a corresponding number of medical scientific research documents for reference by scientific researchers, and can also accurately find a certain medical scientific research document. To further effectively solve the time-consuming search of medical scientific research literature, it is necessary to edit professional vocabulary and then combine the professional vocabulary before searching. If the professional vocabulary cannot form an effective search formula and the literature you want to query cannot be matched, the impact will be Questions about researchers’ experience with the system.

The retrieval method based on medical scientific research information of the present invention can be applied to one or more retrieval terminals. The retrieval terminal is a type that can automatically perform numerical calculations and/or information according to preset or stored instructions. The hardware of processing equipment includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

The retrieval terminal can be any electronic product that can perform human-computer interaction with the user, such as a personal computer, tablet computer, smart phone, personal digital assistant (Personal Digital Assistant, PDA), interactive network television (Internet Protocol Television, IPTV), etc.

The network where the retrieval terminal is located includes but is not limited to the Internet, wide area network, metropolitan area network, local area network, virtual private network (Virtual Private Network, VPN), etc.

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Please refer to Figure 1 which is a flow chart of a retrieval method based on medical scientific research information in a specific embodiment. The method includes:

S101. Obtain the search information input by the user;

According to embodiments of the present application, users can input information to be retrieved through corresponding devices such as keyboards, keypads, switches, dials, mice, trackballs, voice recognizers, etc. Of course it’s not limited to typing and speech. It can be entered in the form of sentences or phrases.

S102. The natural language processing method based on large models analyzes the retrieval information, analyzes the sentence structure, word meaning relationships and contextual information, and extracts keyword information;

According to the embodiments of this application, the natural language processing method of large models can use T5, GLM, and GPT models. These models have powerful natural language understanding capabilities. The large model can understand the natural language input by the user, extract the key information, and convert it into a retrieval expression that the retrieval system can understand.

This embodiment uses the semantic understanding ability of the large model to automatically convert the search information input by the user into keyword information, and then matches the corresponding search formula, thereby realizing automatic retrieval. Return the retrieved documents to the user, so that the user can directly interact with the retrieval system using natural language, without manually constructing retrieval expressions through AND or NOT, improving the operability of the retrieval process and improving retrieval efficiency.

As an example, the large model of this embodiment can use Transformer-encoder, Transformer-decoder and Transformer structures, where Transformer-encoder is mainly used to encode input sequences, such as natural language text. Large models consist of multiple identical encoder layers, each containing a self-attention mechanism and a feedforward neural network.

The mathematical formula of the self-attention mechanism is as follows:

Attention(Q1, K1, V1) = softmax(Q1×K1^T 1/ sqrt(d_k1)) * V1;

Among them, Q1, K1 and V1 represent the input matrix of query, key and value respectively, and d_k1 represents the dimension of the attention mechanism.

The mathematical formula of feedforward neural network is as follows:

FFN(x1) = max(0, xW_1 + b_1)W_2 + b_2;

Among them, x1 represents the input vector, and W_1, b_1, W_2 and b_2 represent the parameters of the model.

For the Transformer-decoder structure, it is mainly used to generate output sequences, such as machine translation tasks. The large model consists of multiple identical decoder layers, each of which contains a self-attention mechanism and an encoder-decoder attention mechanism. The mathematical formula of the self-attention mechanism is the same as that in Transformer-encoder.

The mathematical formula of the encoder-decoder attention mechanism in this embodiment is as follows:

Attention(Q2, K2, V2) = softmax(Q2×K2^T 2/ sqrt(d_k2)) * V2;

Among them, Q2 represents the query vector of the decoder, K2 represents the key vector of the encoder, and V2 represents the value vector of the encoder. The mathematical formula of feedforward neural network is the same as that in Transformer-encoder.

For Transformer, the Transformer structure combines Transformer-encoder and Transformer-decoder for sequence-to-sequence tasks, such as machine translation. It consists of multiple encoder layers and multiple decoder layers stacked alternately.

In the Transformer structure, the connection between the encoder layer and the decoder layer uses residual connection and layer normalization.

The mathematical formula of the residual connection is as follows:

LayerNorm(x3+ Sublayer(x3));

Among them, x3 represents the input vector, and Sublayer(x3) represents the output of the sublayer.

The mathematical formula for layer normalization is as follows:

LayerNorm(x3) = (x3 - mean(x3)) / sqrt(var(x3) + Ep) * GA + Be;

Among them, mean(x3) and var(x3) represent the mean and variance of x3 respectively, Ep is a small constant for numerical stability, and GA and Be are learnable parameters.

Through the above processing method, the Transformer model can effectively capture the contextual information in the input sequence and generate the corresponding output sequence.

In an exemplary embodiment, training of a large model generally includes the following steps:

Data preparation: Prepare large-scale training data sets, including input samples and corresponding target outputs.

Model initialization: Initialize the weights and bias parameters of the neural network.

Forward propagation: Forward propagation of input samples through the neural network to obtain the predicted output of the model.

Calculate the loss: Compare the model's predicted output to the target output, calculate the value of the loss function, and measure the difference between the predicted results and the true results.

Backpropagation: Through the backpropagation algorithm, the gradient of the loss function to the model parameters is calculated, that is, the derivative of the loss function to the weights and biases.

Parameter update: Use an optimization algorithm (such as gradient descent) to update the parameters of the model based on gradient information so that the loss function gradually decreases.

Repeat iteration: Repeat the steps of forward propagation, loss calculation, back propagation, and parameter update until a predetermined number of training epochs or convergence conditions are reached.

The training process of large models usually requires the use of large-scale computing resources and training time. Distributed training, parallel computing and other technologies can be used to accelerate the training process.

In this embodiment, the large model learns the statistical laws and semantic information of language by exposing it to a large amount of natural language text data during the training process. It learns rich language knowledge by modeling the relationships between words, sentences and context in text data. When the user inputs natural language, the large model can understand and process the input based on the learned knowledge, extract the key information, and convert it into a retrieval expression that the retrieval system can understand.

According to embodiments of the present application, the method uses natural language processing technology to perform semantic analysis and understanding of the user's natural language input. By analyzing the sentence structure, word meaning relationships and contextual information input by the user, the system can accurately understand the user's retrieval intention and extract key information. For example, the user can enter "find new drugs to treat cancer", and the system will understand that the user needs to find new drugs related to the treatment of cancer. Here, the information input by the user is split. It can be split based on verbs and nouns, and it can be split into attributives, adverbials, etc. Then identify the keywords and combine them to form a search formula for searching. For "finding new drugs to treat cancer", it can be split into "treating cancer" and "drugs". This matches the corresponding documents. For "cancer treatment" and "drugs", the logical combination method can be based on the user's selection. For example, if the user selects and logically, "cancer treatment" and "drugs" are logically combined to form a search combination to meet the search requirements.

The overall process steps of the natural language processing method of the present invention usually include the following stages:

Word segmentation: segment the input natural language text into words.

Part-of-speech tagging: determine the part of speech (such as noun, verb, etc.) for each word.

Syntactic analysis: Analyze the structure of sentences and determine the dependencies between words.

Semantic analysis: Understand the semantics of sentences, determine the meaning of sentences and the meaning of expressions.

Semantic understanding: extract key information and semantic roles from sentences, and understand the meaning and intention of the sentences.

In the process of semantic analysis and understanding, commonly used mathematical models include:

Word embedding models (such as Word2Vec, GloVe): map words to continuous vector space and capture the semantic relationships between words.

Recurrent Neural Network (RNN): used to process sequence data, such as sentences or text.

Attention Mechanism: used to give weighted attention to different parts of the input when processing long text.

Transformer: A neural network model based on the self-attention mechanism for processing sequence data.

In this embodiment, after analyzing the retrieval information input by the user, the sentence structure and word meaning relationship can be formed. Natural Language Processing (NLP) technology and corresponding mathematical models can usually be used here.

Here's one way the process is done with a large model:

Lexical Analysis: The large model first performs lexical analysis on the sentences input by the user and divides the sentences into words. This can be achieved using a lexer or a pre-trained lexer model.

Syntactic Analysis: Next, the large model performs syntactic analysis, analyzes the structure of the sentence, and determines the dependency relationships between words, such as subject-predicate relationships, verb-object relationships, etc. Parsing can be accomplished using a dependency parser, a phrase structure parser, or a pre-trained parsing model.

Semantic Analysis: In the semantic analysis stage, the large model understands the semantics of the sentence and determines the meaning of the sentence and the meaning of the expression. This includes tasks such as word sense disambiguation, reference resolution, and semantic role annotation. Semantic analysis can be implemented using word sense disambiguation models, reference resolution models, semantic role annotation models, or pre-trained semantic analysis models.

Context Modeling: In order to better understand the meaning of a sentence, the large model considers the contextual information in the sentence, including the preceding and following context. Context modeling can use context-aware models, such as recurrent neural networks (RNN) or transformer models (Transformer), to model and represent sentences.

Through the above steps, the large model can analyze the sentences input by the user and extract the structure, word meaning relationship and contextual information of the sentences, so as to better understand the user's intentions and needs. In this way, the large model can further convert user input into retrieval expressions that the retrieval system can understand and process, and provide more accurate retrieval results. It should be noted that the specific implementation methods and models adopted may vary depending on application scenarios and specific requirements.

S103. Obtain the logical operators set by the user, combine the extracted keyword information based on the logical operators set by the user, and configure it into a search expression;

The search expression in this embodiment is formed by effectively combining the extracted keyword information.

It should be noted that the combination method of keyword information can be obtained by using logical operators set by the user. Based on the logical operators set by the user, the extracted keyword information is combined to form a search expression.

Optionally, logical operators include: AND logic, OR logic, and non-logic.

For example, when searching for cancer drugs, users can select logical relationships while inputting the search information.

Of course, the system can also default the logical relationship to AND logic. That is "treatment" and "cancer" and "drug". Of course, it can also be set to OR logic.

In this way, the search formula is formed by using natural language processing to convert the natural language input by the user into a structured search expression so that the search system can understand and process it.

S104. Match the medical scientific research literature in the medical scientific research literature database based on the search expression;

It is understandable that a large amount of medical scientific research literature is pre-stored and collected in the medical scientific research literature database. Of course, it is not limited to medical scientific research literature and can also involve other literature. In order to facilitate the systematic search of documents, each medical scientific research document is configured with a keyword tag; of course, if the document involves many chapters, multiple keyword tags can be configured on the medical scientific research document, and only the matching search keywords will be used. The document can be matched.

The setting of keyword tags can be set based on the subject name, involved field, abstract, core chapter content, etc. of medical scientific research documents. It can be automatically matched by the system or manually set.

In this way, the keyword tags in the medical scientific research literature database are matched based on the search expression; the medical scientific research documents corresponding to the matched keyword tags are displayed on the user interface.

S105. Display the matched medical scientific research documents on the user interface.

According to the embodiments of the present application, the present invention provides a user-friendly interface, and the user can input search requirements through natural language expression. The user interface can also provide real-time feedback to help users better understand and adjust retrieval needs.

Here, the search results can be continuously optimized based on user feedback, which can be done in one of the following ways: Feedback loop: After the user obtains the search results, the system can ask the user to provide feedback, such as through ratings, likes/dislikes, etc. Based on user feedback, the system can adjust the search algorithm or reorder the results to provide results that better suit the user's needs.

Reinforcement learning: The system can use reinforcement learning algorithms to learn how to optimize search results through interaction with users. The system adjusts the parameters of the retrieval algorithm based on user feedback (reward signals) so that future retrieval results are more in line with user needs.

User model: The system can establish a user model based on the user's historical behavior and preferences, use the model to predict the user's preferences and needs, and optimize the search results based on the prediction results.

It should be understood that the sequence number of each step in the above embodiment does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

It can be seen that in the retrieval method based on medical scientific research information involved in the present invention, the user inputs retrieval requirements through natural language expression, and the interface passes the user input to the large model for natural language understanding. The large model converts user input into retrieval expressions and passes them to the retrieval system. The retrieval system retrieves relevant medical research documents and data from the database according to the search expression, and returns the results to the user interface to display to the user. In this way, users do not need to edit complex search expressions and can search through natural language expressions, reducing users' learning costs and editing time. Moreover, the large model involved in the present invention has natural language understanding capabilities, can more accurately understand the user's intention, and improve the relevance of the search results. The user-friendly interface provides a better user experience, making retrieval easy even for non-experts.

The following is an example of a retrieval system based on medical and scientific research information provided by the embodiment of the present disclosure. This system and the retrieval method based on medical and scientific research information in the above embodiments belong to the same inventive concept. In the implementation of the retrieval system based on medical and scientific research information For details not described in detail in the example, please refer to the above embodiment of the retrieval method based on medical scientific research information.

As shown in Figure 2, the system includes: information input module, information analysis module, expression configuration module, document matching module and document display module;

The information input module provides the user with an input device for information input, and obtains the search information input by the user based on the input device;

The information parsing module parses the retrieved information based on the natural language processing method of large models, parses out the sentence structure, word meaning relationships and contextual information, and extracts keyword information;

The expression configuration module is used to configure the extracted keyword information into a search expression;

The literature matching module matches the medical research literature in the medical research literature database based on the search expression;

The literature display module provides a display module for system operation information, displays retrieval process information, and displays matched medical scientific research documents on the user interface.

The retrieval system based on medical scientific research information provided by the present invention is based on the units and algorithm steps of each example described in the embodiments disclosed herein, and can be implemented by electronic hardware, computer software, or a combination of both. In order to clearly illustrate the hardware and software interchangeability. In the above description, the composition and steps of each example have been generally described according to functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may implement the described functionality using different methods for each specific application, but such implementations should not be considered to be beyond the scope of the present invention.

Through the above description of the embodiments, those skilled in the art can easily understand that the retrieval system based on medical scientific research information described here can be implemented by software, or can be implemented by software combined with necessary hardware. Therefore, the technical solution according to the disclosed embodiment of the retrieval method based on medical scientific research information can be embodied in the form of a software product, and the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk etc.) or on a network, including several instructions to cause a computing device (which may be a personal computer, a server, a mobile terminal, a network device, etc.) to execute an indexing method according to an embodiment of the present disclosure.

In embodiments of the present invention, computer program code for performing operations of the present disclosure may be written in one or more programming languages, including but not limited to object-oriented programming languages, or a combination thereof— Such as Java, Smalltalk, C++, but also conventional procedural programming languages - such as "C" language or similar programming languages.

The above description of the disclosed embodiments enables those skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be practiced in other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (9)

1. A retrieval method based on medical scientific research information, characterized in that the method includes: Step 1. Obtain the search information input by the user; Step 2: Analyze the retrieval information based on the natural language processing method of the large model, analyze the sentence structure, word meaning relationship and contextual information, and extract the keyword information; Among them, the large model includes: multiple identical encoder layers, each encoder layer contains a self-attention mechanism and a feed-forward neural network; The mathematical formula of the self-attention mechanism is as follows: Attention(Q1, K1, V1) = softmax(Q1×K1^T 1/ sqrt(d_k1)) * V1; Among them, Q1, K1 and V1 represent the input matrix of query, key and value respectively, and d_k1 represents the dimension of the attention mechanism; The mathematical formula of feedforward neural network is as follows: FFN(x1) = max(0, xW_1 + b_1)W_2 + b_2; Among them, x1 represents the input vector, W_1, b_1, W_2 and b_2 represent the parameters of the model; The large model also includes: multiple identical decoder layers, each of which contains a self-attention mechanism and an encoder-decoder attention mechanism; In the large model structure, the connection between the encoder layer and the decoder layer uses residual connection and layer normalization processing; Step 3: Obtain the logical operators set by the user, combine the extracted keyword information based on the logical operators set by the user, and configure it into a search expression; Step 4: Match the medical research literature in the medical research literature database based on the search expression; Step 5: Display the matched medical research documents on the user interface. 2. The retrieval method based on medical scientific research information according to claim 1, characterized in that the method of parsing the retrieval information based on the natural language processing method of the large model in step 2 also includes: Segment the input natural language text into words; Determine the part of speech of each word and extract keyword information. 3. The retrieval method based on medical scientific research information according to claim 1, characterized in that the method of parsing the retrieval information based on the natural language processing method of the large model in step 2 also includes: Analyze the sentence structure in the retrieved information, determine the dependencies between words, and extract keyword information. 4. The retrieval method based on medical scientific research information according to claim 3, characterized in that the method of parsing the retrieval information based on the natural language processing method of the large model in step 2 also includes: parsing the retrieval information input by the user. Sentences are lexically analyzed and divided into words. 5. The retrieval method based on medical scientific research information according to claim 4, characterized in that performing lexical analysis on the sentence includes: determining the dependency relationship between words in the sentence, and using dependency syntax analysis or phrase structure syntax analysis. way to achieve it. 6. The retrieval method based on medical scientific research information according to claim 1, characterized in that logical operators include: AND logic, OR logic and non-logic. 7. The retrieval method based on medical scientific research information according to claim 1, characterized in that step 4 further includes: storing multiple medical scientific research documents in the medical scientific research document database; Each medical research document is configured with keyword tags; Match the keyword tags in the medical research literature database based on the search expression; The medical scientific research documents corresponding to the matched keyword tags are displayed on the user interface. 8. A retrieval system based on medical scientific research information, characterized in that the system implements the retrieval method based on medical scientific research information as described in any one of claims 1 to 7; The system includes: information input module, information analysis module, expression configuration module, document matching module and document display module; The information input module is used to obtain the retrieval information input by the user; The information parsing module is used to parse the retrieved information in combination with the natural language processing method of the large model, parse out the sentence structure, word meaning relationships and contextual information, and extract keyword information; Among them, the large model includes: multiple identical encoder layers, each encoder layer contains a self-attention mechanism and a feed-forward neural network; The mathematical formula of the self-attention mechanism is as follows: Attention(Q1, K1, V1) = softmax(Q1×K1^T 1/ sqrt(d_k1)) * V1; Among them, Q1, K1 and V1 represent the input matrix of query, key and value respectively, and d_k1 represents the dimension of the attention mechanism; The mathematical formula of feedforward neural network is as follows: FFN(x1) = max(0, xW_1 + b_1)W_2 + b_2; Among them, x1 represents the input vector, W_1, b_1, W_2 and b_2 represent the parameters of the model; The large model also includes: multiple identical decoder layers, each of which contains a self-attention mechanism and an encoder-decoder attention mechanism; In the large model structure, the connection between the encoder layer and the decoder layer uses residual connection and layer normalization processing; the expression configuration module is used to obtain the user settings in the logical operator, and based on the user set logic The operator combines the extracted keyword information and configures it into a search expression; The literature matching module is used to match the medical research literature in the medical research literature database based on the search expression; The document display module is used to display the search process information and display the matched medical scientific research documents. 9. A retrieval terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the processor executes the program, the processor implements claim 1 Go to the steps of the retrieval method based on medical scientific research information described in any one of 7.