Medical literature automatic classification system and semantic retrieval method based on deep learning

Abstract

The invention provides a medical literature automatic classification system and semantic retrieval method based on deep learning, and belongs to the technical field of big data processing. Comprising. The data preprocessing module is mainly used for performing cleaning, normalization and feature extraction processing on texts, images and structured data of medical literatures; the deep classification model is used for training historical data based on a mixed neural network architecture fused with a medical knowledge graph, and automatically classifying medical literatures through a multi-stage classification module and a literature utilization rate analysis module; the multi-level classification module is mainly used for performing hierarchical primary classification on medical literatures according to predefined medical subject large categories and sub-categories thereof; and the literature utilization rate analysis module mainly constructs a dynamic weight model based on the related data of the literature, and classifies and sorts each sub-class literature based on the model data. The method has the advantages that automatic classification and semantic retrieval of medical literatures are achieved, and classification precision and retrieval efficiency are improved through multi-level classification and literature utilization rate analysis.

Description

Technical Field

[0001] The present invention relates to the field of big data processing technology, and in particular to an automatic classification system and semantic retrieval method for medical literature based on deep learning. Background Art

[0002] With the rapid development of medical research, the volume of medical literature worldwide has increased dramatically. Medical literature contains a vast array of specialized terminology, complex medical concepts, and multidimensional information, making manual classification and retrieval not only time-consuming but also prone to omissions. Consequently, automated classification and semantic retrieval methods for medical literature have emerged to improve the efficiency of document management and knowledge discovery. By performing semantic analysis on the content of medical literature, it is possible to automatically categorize documents into corresponding topics or fields, significantly improving the efficiency of document organization and access.

[0003] Current automatic classification systems and semantic retrieval methods for medical literature on the market generally suffer from insufficient accuracy and efficiency. Traditional classification methods often rely on simple keyword matching or rule-based algorithms, making it difficult to accurately understand the deep semantics of medical literature, resulting in low accuracy in classification results. Many existing systems fail to effectively incorporate medical expertise and ignore the deep integration of medical knowledge graphs, causing classification models to be biased when processing complex documents and unable to adapt to the diversity and complexity of document content. Traditional methods rely more on word frequency-based retrieval technology, lack a deep understanding of medical terminology, and are unable to truly capture the user's query intent and the semantic relationship between documents, resulting in inaccurate retrieval results. Existing systems generally fail to dynamically adjust the weight and influence of documents, ignoring the impact of data such as document usage frequency and citation volume. They are unable to reflect the actual influence and popularity of documents in real time, affecting the rationality of document ranking. Summary of the Invention

[0004] In order to improve the existing automatic classification system and semantic retrieval method of medical literature, this paper provides an automatic classification system and semantic retrieval method of medical literature based on deep learning. This method uses deep learning and medical knowledge graphs to realize automatic classification and semantic retrieval of medical literature, improves classification accuracy and retrieval efficiency through multi-level classification and literature usage analysis, and can accurately match user queries with relevant literature.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: An automatic classification system for medical literature based on deep learning, including: Data preprocessing module: The data preprocessing module is mainly used for cleaning, normalizing and feature extraction of text, images and structured data of medical literature; Deep classification model: The deep classification model is based on a hybrid neural network architecture integrated with a medical knowledge graph to train historical data. It automatically classifies medical literature through a multi-level classification module and a literature usage analysis module. Multi-level classification module: The multi-level classification module is mainly used to perform hierarchical initial classification of medical literature according to predefined medical discipline categories and their subcategories; Document usage analysis module: The document usage analysis module mainly builds a dynamic weight model based on the document's citation frequency, download volume, and access time series data, and classifies and sorts each subcategory of documents based on the model data.

[0006] Preferably, the deep classification model specifically includes: Medical knowledge graph fusion unit: Based on the medical literature data obtained by the data preprocessing module, the medical knowledge graph fusion unit extracts the relationship between medical entities through natural language processing, converts the entities and relationships in the medical knowledge graph into embedding vectors, and constructs a knowledge graph; Model architecture unit: The model architecture unit is mainly used to encode the text of medical documents through a pre-trained language model, and fuse the embedding vector obtained from the medical knowledge graph with the text encoding result. The fused information is further processed through the MLP layer to extract the high-level semantic features of the document.

[0007] Preferably, the multi-level classification module specifically includes: Large category classification unit: The large category classification unit is mainly used to classify all documents belonging to the same large category into the same area based on the high-level semantic features of the acquired medical documents. Small category classification unit: The small category classification unit is based on the large category classification, and further classifies medical documents belonging to the same large category into more specific subcategories; Label propagation unit: The label propagation unit is mainly used to clarify the relationship between hierarchical classification classes. The classification results of each layer will affect the classification tasks of subsequent layers to ensure the consistency of classification.

[0008] Preferably, the document usage analysis module specifically includes: Feature extraction unit: The feature extraction unit is mainly used to extract key features from the usage data of the document, including citation frequency, download volume, visit volume and time window; Dynamic weight model building unit: The dynamic weight model building unit assigns a weight to each feature based on the result of feature extraction, and dynamically adjusts the usage data of the document based on the time series; Document classification and ranking unit: The document classification and ranking unit ranks the medical documents of each major category and each minor category based on the document usage rate obtained by the dynamic weight model, and divides them into high-impact documents, medium-impact documents, and low-impact documents.

[0009] Furthermore, the semantic retrieval method of medical literature based on deep learning includes: Parse user query statements into semantic vectors based on a language model pre-trained in the medical field; Generate an expanded query based on the search context, and calculate the cosine similarity between the query vector and the document semantic vector library based on the query vector; Based on cosine similarity, the ranking results are obtained, and the semantic association map between documents is output simultaneously to show the knowledge evolution path of core concepts.

[0010] Preferably, the parsing of user query statements into semantic vectors based on a language model pre-trained in the medical field specifically includes: Perform medical terminology recognition and standardization on pre-processed user query sentences; Based on a large pre-trained language model in the medical field, the processed user query sentences are input into the model for encoding, and the corresponding semantic vector representation is generated based on the position and context of each word in the sentence; A vector representing the entire query statement is extracted from the vector of each token output by the model through the pooling method to obtain the semantic vector.

[0011] Preferably, generating an extended query in combination with the search context and calculating the cosine similarity between the query vector and the document semantic vector library specifically includes: Based on the user's initial query and context, an expanded query is generated to obtain more relevant terms, domain knowledge, and contextual information; Based on the obtained semantic vector, for each document, perform a dot product operation between its semantic vector and the query vector; Calculate the modulus of the semantic vector and each document vector; The cosine similarity between the semantic vector and each document vector is calculated using the formula.

[0012] Preferably, obtaining the ranking results based on cosine similarity and synchronously outputting the semantic association map between documents to display the knowledge evolution path of the core concepts specifically includes: Based on the results of cosine similarity calculation, the documents are sorted and the list of documents most relevant to the expanded query is returned; Based on the similarity and common semantic features between documents, a relationship graph between documents is constructed. In the graph, core concepts are displayed as nodes, and nodes are connected by edges to represent the relationship between them. Based on the citation relationship between documents and the temporal evolution of concepts, the knowledge evolution path of core concepts is obtained.

[0013] Compared with the prior art, the advantages of the present invention are: Through advanced natural language processing and deep learning technologies, it provides an efficient and accurate solution for document management and retrieval in the medical field. By integrating medical knowledge graphs with hybrid neural network architectures, the system can deeply understand the content and semantics of medical documents, automatically perform multi-level classification and hierarchical management, and effectively improve the accuracy and efficiency of document classification. The document usage analysis module combines data such as citation frequency and download volume to dynamically adjust document weights and optimize the sorting and classification of documents. In addition, the semantic retrieval method accurately matches user queries with relevant content in the document library through pre-trained language models and cosine similarity calculations, and generates semantic association maps between documents, showing the knowledge evolution path of core concepts, further promoting the efficient retrieval and application of documents. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which are incorporated herein and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, further serve to explain the principles of the invention and to enable one skilled in the art to make and use the invention.

[0015] Figure 1 A schematic diagram of the system proposed by the present invention; Figure 2 A schematic diagram of the method proposed in the present invention; Figure 3 This is the deep classification model diagram proposed by the present invention; Figure 4 This is the multi-level classification module diagram proposed by the present invention; Figure 5 This is a module diagram for analyzing the literature usage rate proposed in the present invention; Figure 6 This is a schematic diagram of obtaining the semantic vector proposed by the present invention; Figure 7 This is a schematic diagram of the cosine similarity calculation proposed by the present invention; Figure 8 This is a schematic diagram of the sorting results proposed by the present invention; Figure 9 This is a diagram of the architecture of the electronic equipment in this solution; Figure 10 This is a schematic diagram of the computer-readable storage medium structure in this solution.

[0016] As shown in the figure, in order to clearly implement the structure of the embodiment of the present invention, specific structures and devices are marked in the figure, but this is only for illustrative purposes and is not intended to limit the present invention to the specific structure, device and environment. According to specific needs, ordinary technicians in this field can adjust or modify these devices and environments. DETAILED DESCRIPTION

[0017] The following describes in detail the deep learning-based automatic classification system and semantic retrieval method for medical literature provided by the present invention, with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative implementations for certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0018] It should be noted that references in the specification to "one embodiment," "an embodiment," "exemplary embodiments," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment will include such specific features, structures, or characteristics. Furthermore, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0019] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0020] It will be understood that the meanings of “on,” “over,” and “above” in the present invention should be interpreted in the broadest manner, so that “on” means not only “directly on” something but also includes the meaning of being “on” something with intervening features or layers, and “on” or “above” means not only “on” or “above” something but also includes the meaning of being “on” or “above” something with no intervening features or layers.

[0021] Additionally, spatially relative terms such as "below," "beneath," "lower," "above," and "upper" may be used herein for descriptive convenience to describe the relationship of one element or feature to another element or features, as illustrated in the accompanying drawings. Spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the accompanying drawings. The device may be oriented in other ways, and the spatially relative descriptors used herein should be similarly interpreted accordingly.

[0022] See Figure 1 As shown in the figure, the automatic classification system of medical literature based on deep learning includes: Data preprocessing module: The data preprocessing module is mainly used for cleaning, normalizing and feature extraction of text, images and structured data of medical literature; Deep classification model: The deep classification model is based on a hybrid neural network architecture integrated with a medical knowledge graph to train historical data. It automatically classifies medical literature through a multi-level classification module and a literature usage analysis module. Multi-level classification module: The multi-level classification module is mainly used to perform hierarchical initial classification of medical literature according to predefined medical discipline categories and their subcategories; Document usage analysis module: The document usage analysis module mainly builds a dynamic weight model based on the document's citation frequency, download volume, and access time series data, and classifies and sorts each subcategory of documents based on the model data.

[0023] See Figure 2 As shown in Figure 2, the semantic retrieval method for medical literature based on deep learning includes: Parse user query statements into semantic vectors based on a language model pre-trained in the medical field; Generate an expanded query based on the search context, and calculate the cosine similarity between the query vector and the document semantic vector library based on the query vector; Based on cosine similarity, the ranking results are obtained, and the semantic association map between documents is output simultaneously to show the knowledge evolution path of core concepts.

[0024] See Figure 3 As shown in Figure 2, the deep classification model specifically includes: Medical knowledge graph fusion unit: Based on the medical literature data obtained by the data preprocessing module, the medical knowledge graph fusion unit extracts the relationship between medical entities through natural language processing, converts the entities and relationships in the medical knowledge graph into embedding vectors, and constructs a knowledge graph; Model architecture unit: The model architecture unit is mainly used to encode the text of medical documents through a pre-trained language model, and fuse the embedding vector obtained from the medical knowledge graph with the text encoding result. The fused information is further processed through the MLP layer to extract the high-level semantic features of the document.

[0025] Specifically, natural language processing technology is used to extract various medical entities and their relationships from medical literature. Entities may include diseases, symptoms, drugs, treatment methods, etc., and relationships may include "treatment", "cause", "prevention", etc. Based on the extracted entities and relationships, a medical knowledge graph is constructed, which is represented as: ,in Representing an entity and There is a relationship between ,The edges in the graph represent different types of relationships; Convert entities and relations in the medical knowledge graph into embedding vectors, typically using graph neural networks or graph-based embedding methods to obtain low-dimensional representations of each entity and relation; The embedding vectors in the medical knowledge graph are fused with the text encoding results. The fusion of the embedding vectors can be achieved through simple connections. The fused embedding vectors are input into the multi-layer perceptron (MLP) for further processing to extract the high-level semantic features of the document. The MLP can learn more complex feature representations through multi-layer nonlinear transformations.

[0026] See Figure 4 As shown, the multi-level classification module specifically includes: Large category classification unit: The large category classification unit is mainly used to classify all documents belonging to the same large category into the same area based on the high-level semantic features of the acquired medical documents. Small category classification unit: The small category classification unit is based on the large category classification, and further classifies medical documents belonging to the same large category into more specific subcategories; Label propagation unit: The label propagation unit is mainly used to clarify the relationship between hierarchical classification classes. The classification results of each layer will affect the classification tasks of subsequent layers to ensure the consistency of classification.

[0027] Specifically, the input large-category classification unit is the high-level semantic features of medical documents extracted by the previous model architecture unit. The high-level semantic features are input into the large-category classification unit to predict the large category. The prediction result of the large category represents the large category to which the document belongs. Based on the classification result, the document is divided into the corresponding large category area. The task of the sub-category classification unit is to further classify medical documents belonging to the same major category into more refined sub-categories based on the major category classification. The input of the sub-category classification unit is the high-level semantic features obtained after the major category classification, and its output is the sub-category label to which the document belongs. In a multi-level classification system, the classification results of each level should be consistent and have a clear hierarchical relationship between each other. The label propagation unit ensures classification consistency by providing feedback on the classification results of each level. Based on the classification results of each layer, the category labels are propagated from the large category layer to the small category layer. Specifically, the results of the small category classifier will be affected by the large category classification results, ensuring that the documents under the same large category meet the consistency constraints when classified into small categories; The label propagation unit ensures that the classification results of each layer have an impact on the classification of subsequent layers to achieve consistency in the overall classification process. Through this mechanism, it can be ensured that the final classification results conform to the hierarchical structure of the entire multi-level classification.

[0028] See Figure 5 As shown in the figure, the literature usage analysis module specifically includes: Feature extraction unit: The feature extraction unit is mainly used to extract key features from the usage data of the document, including citation frequency, download volume, visit volume and time window; Dynamic weight model building unit: The dynamic weight model building unit assigns a weight to each feature based on the result of feature extraction, and dynamically adjusts the usage data of the document based on the time series; Document classification and ranking unit: The document classification and ranking unit ranks the medical documents of each major category and each minor category based on the document usage rate obtained by the dynamic weight model, and divides them into high-impact documents, medium-impact documents, and low-impact documents.

[0029] Specifically, the key features of the document usage data, including citation frequency, download volume, visit volume and time window, are obtained through the feature extraction unit to form a feature vector ,in is the feature vector of the i-th document, is the citation frequency, For downloads, For visits, is the time window; Assign an initial weight to each feature: ,in The corresponding citation frequency, Corresponding download volume, Corresponding visits, Corresponding time window; The weights are dynamically adjusted according to the time series data. The literature usage data changes over time, and the weights of the features can be adjusted by the time decay function. The decay factor formula is:

[0030] in, is the attenuation factor, is a hyperparameter that controls the influence of time. For time; The feature weights after dynamic adjustment are: , is the dynamic weight at time t; According to the dynamic weight model, the characteristics of each document are weighted with its dynamic weight. The weighted feature vector of each document is:

[0031] in, is the document feature vector after dynamic weight adjustment; According to the weighted eigenvector , calculate the usage rate of the literature , the formula is:

[0032] in, is the value of document i on the jth feature; Based on calculated usage , sort the literature, and the sorted literature can be divided into three categories: high influence, medium influence and low influence.

[0033] See Figure 6 As shown in the figure, parsing user query statements into semantic vectors based on the language model pre-trained in the medical field specifically includes: Perform medical terminology recognition and standardization on pre-processed user query sentences; Based on a large pre-trained language model in the medical field, the processed user query sentences are input into the model for encoding, and the corresponding semantic vector representation is generated based on the position and context of each word in the sentence; A vector representing the entire query statement is extracted from the vector of each token output by the model through the pooling method to obtain the semantic vector.

[0034] Specifically, the standardized medical terminology includes: uniformly mapping the identified medical terminology to standardized terminology or code through a standardized mapping table; Spelling correction of terms to ensure the accuracy of medical terminology and avoid recognition bias caused by different spellings; Through the medical knowledge graph or domain-specific models, further check whether the standardized terms are consistent with the medical context to avoid ambiguity; The pre-trained language model encodes the query using a Transformer-based architecture. The position and context of each word in the sentence are taken into account to generate a context-sensitive embedding vector for each word. In order to obtain the semantic information of the entire query statement, a pooling method can be used. The pooling method usually aggregates the embedding vector of each word to generate a single vector representing the entire query statement. A common pooling method is to take the average of the embedding vectors of each token.

[0035] See Figure 7 As shown in the figure, an extended query is generated based on the search context, and the cosine similarity between the query vector and the document semantic vector library is calculated based on the query vector. Specifically, the following steps are performed: Based on the user's initial query and context, an expanded query is generated to obtain more relevant terms, domain knowledge, and contextual information; Based on the obtained semantic vector, for each document, perform a dot product operation between its semantic vector and the query vector; Calculate the modulus of the semantic vector and each document vector; The cosine similarity between the semantic vector and each document vector is calculated using the formula.

[0036] Specifically, for each document i, its semantic vector is expressed as , by calculating the query vector and the semantic vector of each document The dot product of is used to measure the similarity between the query and the document. The dot product calculation formula is:

[0037] in, is the j-th dimension of the query statement, is the j-th dimension of document i, and m is the dimension of the vector; To normalize the similarity, the query vector is calculated and literature vector The modulus length is 1, which avoids the influence of different vector lengths on similarity calculation; Based on the calculated dot product and modulus, the cosine similarity is calculated to measure the similarity between the query vector and the document vector.

[0038] See Figure 8 As shown in the figure, based on cosine similarity, the ranking results are obtained, and the semantic association map between documents is output simultaneously, showing the knowledge evolution path of the core concepts. Specifically, it includes: Based on the results of cosine similarity calculation, the documents are sorted and the list of documents most relevant to the expanded query is returned; Based on the similarity and common semantic features between documents, a relationship graph between documents is constructed. In the graph, core concepts are displayed as nodes, and nodes are connected by edges to represent the relationship between them. Based on the citation relationship between documents and the temporal evolution of concepts, the knowledge evolution path of core concepts is obtained.

[0039] Specifically, based on the calculated cosine similarity, the documents are sorted in descending order of similarity value. The list of documents most relevant to the query is returned; The relationship between documents can be constructed by their semantic similarity and common concepts to form a relationship graph. The nodes of the graph represent the core concepts of the documents, and the edges represent the semantic relationships between these concepts. Based on the citation relationship of the literature and the temporal evolution of the concepts, the development and evolution path of the core concepts can be tracked. The specific steps include: Obtain citation relationships between documents and construct citation networks; Arrange the literature in chronological order, construct a time series of literature, observe the citation status of literature at different time points, and track the changes in core concepts; By combining citation relationships and time series, we can construct the knowledge evolution path of core concepts. For example, if document A cites document B, and the publication time of document A is after document B, then document A is the knowledge continuation of document B.

[0040] Furthermore, the method according to the embodiment of the present application can also be used with the aid of Figure 9 The electronic device architecture shown in FIG. Figure 9 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store the medical literature automatic classification system and semantic retrieval method based on deep learning provided in this application. The electronic device 500 may also include a terminal interface 508. Of course, Figure 9 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 9 One or more components of an electronic device are shown.

[0041] Figure 10 This is a schematic diagram of the computer-readable storage medium structure provided by an embodiment of the present application. Figure 10As shown, a computer-readable storage medium 600 according to one embodiment of the present application is shown. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by the processor, the medical literature automatic classification system and semantic retrieval method based on deep learning according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0042] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0043] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0044] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

[0045] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0046] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A medical literature automatic classification system based on deep learning, characterized by: include: Data preprocessing module: The data preprocessing module is mainly used for cleaning, normalizing and feature extraction of text, images and structured data of medical literature; Deep classification model: The deep classification model is based on a hybrid neural network architecture integrated with a medical knowledge graph to train historical data. It automatically classifies medical literature through a multi-level classification module and a literature usage analysis module. Multi-level classification module: The multi-level classification module is mainly used to perform hierarchical initial classification of medical literature according to predefined medical discipline categories and their subcategories; Document usage analysis module: The document usage analysis module mainly builds a dynamic weight model based on the document's citation frequency, download volume, and access time series data, and classifies and sorts each subcategory of documents based on the model data.

2. The deep learning-based automatic classification system for medical literature according to claim 1, characterized in that: The deep classification model specifically includes: Medical knowledge graph fusion unit: Based on the medical literature data obtained by the data preprocessing module, the medical knowledge graph fusion unit extracts the relationship between medical entities through natural language processing, converts the entities and relationships in the medical knowledge graph into embedding vectors, and constructs a knowledge graph; Model architecture unit: The model architecture unit is mainly used to encode the text of medical documents through a pre-trained language model, and fuse the embedding vector obtained from the medical knowledge graph with the text encoding result. The fused information is further processed through the MLP layer to extract the high-level semantic features of the document.

3. The deep learning-based automatic classification system for medical literature according to claim 1, characterized in that: The multi-level classification module specifically includes: Large category classification unit: The large category classification unit is mainly used to classify all documents belonging to the same large category into the same area based on the high-level semantic features of the acquired medical documents. Small category classification unit: The small category classification unit is based on the large category classification, and further classifies medical documents belonging to the same large category into more specific subcategories; Label propagation unit: The label propagation unit is mainly used to clarify the relationship between hierarchical classification classes. The classification results of each layer will affect the classification tasks of subsequent layers to ensure the consistency of classification.

4. The deep learning-based automatic classification system for medical literature according to claim 1, characterized in that: The document usage analysis module specifically includes: Feature extraction unit: The feature extraction unit is mainly used to extract key features from the usage data of the document, including citation frequency, download volume, visit volume and time window; Dynamic weight model building unit: The dynamic weight model building unit assigns a weight to each feature based on the result of feature extraction, and dynamically adjusts the usage data of the document based on the time series; Document classification and ranking unit: The document classification and ranking unit ranks the medical documents of each major category and each minor category based on the document usage rate obtained by the dynamic weight model, and divides them into high-impact documents, medium-impact documents, and low-impact documents.

5. A medical literature semantic retrieval method based on deep learning, characterized by: include: Parse user query statements into semantic vectors based on a language model pre-trained in the medical field; Generate an expanded query based on the search context, and calculate the cosine similarity between the query vector and the document semantic vector library based on the query vector; Based on cosine similarity, the ranking results are obtained, and the semantic association map between documents is output simultaneously to show the knowledge evolution path of core concepts.

6. The method for automatic classification of medical literature based on deep learning according to claim 5, characterized in that: The method of parsing the user query statement into a semantic vector based on the language model pre-trained in the medical field specifically includes: Perform medical terminology recognition and standardization on pre-processed user query sentences; Based on a large pre-trained language model in the medical field, the processed user query sentences are input into the model for encoding, and the corresponding semantic vector representation is generated based on the position and context of each word in the sentence; A vector representing the entire query statement is extracted from the vector of each token output by the model through the pooling method to obtain the semantic vector.

7. The method for automatic classification of medical literature based on deep learning according to claim 5, characterized in that: The step of generating an extended query based on the search context and calculating the cosine similarity between the query vector and the document semantic vector library specifically includes: Based on the user's initial query and context, an expanded query is generated to obtain more relevant terms, domain knowledge, and contextual information; Based on the obtained semantic vector, for each document, perform a dot product operation between its semantic vector and the query vector; Calculate the modulus of the semantic vector and each document vector; The cosine similarity between the semantic vector and each document vector is calculated using the formula.

8. The method for automatic classification of medical literature based on deep learning according to claim 5, characterized in that: The method of obtaining the ranking results based on cosine similarity and synchronously outputting the semantic association map between documents to display the knowledge evolution path of the core concepts specifically includes: Based on the results of cosine similarity calculation, the documents are sorted and the list of documents most relevant to the expanded query is returned; Based on the similarity and common semantic features between documents, a relationship graph between documents is constructed. In the graph, core concepts are displayed as nodes, and nodes are connected by edges to represent the relationship between them. Based on the citation relationship between documents and the temporal evolution of concepts, the knowledge evolution path of core concepts is obtained.

9. An electronic device, characterized in that: include: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the deep learning-based medical literature semantic retrieval method as described in any one of claims 5-8.

10. A computer-readable storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by a processor, the medical literature semantic retrieval method based on deep learning as described in any one of claims 5 to 8 is implemented.

Images