Intelligent journal recommendation system and method

Abstract

The invention belongs to the technical field of computer artificial intelligence, and discloses an intelligent periodical recommendation system and method.According to the intelligent periodical recommendation system and method, a traditional deep learning technology and a more advanced large model technology are combined, a periodical list d1 is generated from to-be-recommended paper information through a large model, meanwhile, a periodical list d2 is generated from the to-be-recommended paper information through a deep learning model M1, and the periodical list d1 and the periodical list d2 are integrated. Comparing the periodical list d1 with the periodical list d2 to obtain a periodical list d3, performing periodical verification on the periodical list d3 in combination with the periodical database DB1 to obtain a periodical dictionary d4, and obtaining a periodical recommendation scheme; according to the method, the problems of high requirements on data quality and data volume and low accuracy of predicting and recommending periodicals caused by purely using a deep learning model are avoided, and the prediction accuracy is effectively improved.

Description

Technical Field

[0001] The present application belongs to the field of computer artificial intelligence technology, and specifically relates to an intelligent journal recommendation system and method. Background Art

[0002] At present, journal recommendations are generally based on deep learning technology. Through trained deep learning models, such as LSTM, a mapping relationship between existing papers and journals is established, and then the journals to which new papers should be submitted are predicted; Problems and shortcomings: A There are many multi-classification problems. Using deep learning classification models for prediction makes prediction very difficult, which aggravates the problem of inaccurate prediction. B. Deep learning models have very high requirements for data quality and data volume. Simply using deep learning models to predict the journals to which papers should be submitted may result in inaccurate predictions. C Most existing deep learning models can only be trained offline and cannot obtain the latest journal dynamics and information in a timely manner, which affects the accuracy of prediction; Application Contents The purpose of this application is to solve the problems existing in the prior art and to provide an intelligent journal recommendation system and method.

[0003] In order to solve the technical problem, the technical solution of the present application is: an intelligent journal recommendation method, comprising the following steps: Step 1: Build a journal database DB1, which contains journal data and published paper data; Step 2: Build a crawler module to crawl the latest journal data from the Internet public database and website and save it to the journal database DB1; Step 3: Based on the journal data and published paper data in the journal database DB1 and based on the pytorch framework, train the deep learning model M1; Step 4: Build the ElasticSearch search engine E1 on the journal database DB1; Step 5: The input module receives the information of the paper to be recommended input by the user; Step 6: The input module combines the information of the paper to be recommended with the journal prediction prompt words, and inputs them into the big model through the big model interface. The big model generates a recommended journal list d1, and the journals in d1 are sorted according to the journal recommendation degree given by the big model; Step 7: The input module inputs the information of the papers to be recommended into the deep learning model M1, and generates a recommended journal list d2. The journals in d2 are sorted according to the journal recommendation degree given by the deep learning model M1; Step 8: Compare the journal list d1 and the journal list d2, retain the duplicate journals, delete the non-duplicate journals, and obtain the recommended journal list d3; Step 9: Based on the ElasticSearch search engine E1 and the journal database DB1, the journal information in the journal list d3 is compared with the journal data and published paper data in the journal database DB1 to complete the journal verification. At the same time, the latest journal information updated by the crawler module in the journal database DB1 is added to the journal list d3 to form a journal dictionary d4. Step 10: The data of the journal dictionary d4 is transmitted back to the user end of the journal recommendation system and output to obtain a journal recommendation scheme.

[0004] Preferably, the journal data in step 1 includes journal name, introduction, target readership, subject area, impact factor, publisher, publication cycle, A partition, B partition, publication ratio, review cycle, page charge, and whether it is Open Access; the published paper data includes paper title, abstract, full text, subject area, keywords, subject tags, target journal impact factor, and target partition.

[0005] Preferably, in step 2, the crawler module legally and compliantly crawls the latest journal data from public databases and websites on the Internet, saves it to the journal database DB1, and fully updates it once a week.

[0006] Preferably, the training process of the deep learning model M1 in step 3 includes the following steps: Step 3-1, data preparation: prepare the journal data and published paper data to be trained, and perform data preprocessing so that they can be fully and correctly input into the model; Step 3-2, model design: According to the journal data to be input and the format of published paper data, as well as the format of the journal data to be output, the structure of the deep learning model is designed, including a text embedding module, a domain feature embedding module, a feature fusion module, and an output classification module; The text embedding module includes journal text embedding and paper text embedding; Journal text embedding: Use a pre-trained language model to embed the journal name, introduction, and target readership to obtain the semantic representation of the journal; Paper text embedding: Use the pre-trained language model to embed the paper title, abstract, and keywords to obtain the semantic representation of the paper; The domain feature embedding module includes journal domain feature embedding and paper domain feature embedding; Embedding journal domain features: embedding the journal’s subject field, impact factor, and review cycle to obtain the journal’s domain feature representation; Embedding the field features of the paper: embed the field, keywords, and topic tags of the paper to obtain the field feature representation of the paper; The feature fusion module includes splicing fusion and attention mechanism; Splicing and fusion: Splicing journal text embedding, journal field feature embedding, paper text embedding, and paper field feature embedding to form a comprehensive feature vector; Attention mechanism: Introduce the attention mechanism to assign different weights to different feature vectors and obtain the fused features; The output classification module includes a fully connected layer and a Softmax layer; Fully connected layer: The fused features are processed through a series of fully connected layers to output the matching score of each journal; Softmax layer: Normalizes the output matching scores and converts them into the predicted probability value of each journal, i.e., the recommendation degree. At the same time, the difference between the matching probability of the journal predicted by the model and the actual label is evaluated through the multi-label classification loss function. Step 3-3, model training and optimization: input journal data and published paper data into the deep learning model for training. During the training process, the model performance is evaluated using model evaluation indicators. Based on the evaluation results, the model structure and training strategy are adjusted, and the model parameters and hyperparameters are repeatedly optimized until the model indicators reach the expected standards. Step 3-4, model deployment: deploy the trained deep learning model to the deep learning server and open the interface so that the deep learning model can receive the information of the recommended papers in step 5 and output the predicted journal list d2.

[0007] Preferably, the multi-label classification loss function is defined as:

[0008] in: is the actual label, 0 or 1, indicating whether the paper is suitable for the i-th journal; is the matching probability of the journal predicted by the model.

[0009] Preferably, in step 6, combining the information of the paper to be recommended with the journal prediction prompt word specifically comprises: inserting the title, field and abstract of the information of the paper to be recommended into the curly brackets {} after the prediction prompt word respectively.

[0010] Preferably, the large model in step 6 is the DeepSeek-R1-671B large model, and the recommendation degree refers to the matching degree between each journal in the journal list d1 generated by the DeepSeek-R1-671B large model and the paper to be recommended. The DeepSeek-R1-671B large model generates the recommendation degree while generating the journal list d1.

[0011] Preferably, an intelligent journal recommendation system comprises an input module, a large model module, a deep learning model module, a comparison module, a journal verification module, a journal database module, a crawler module and a journal output module; The input module is used for users to input information about papers to be recommended, combine the information about papers to be recommended with journal prediction prompt words, input the information about papers to be recommended into the big model module through the big model interface, and also input the information about papers to be recommended into the deep learning model module; The large model module is used to generate a recommended journal list d1; The deep learning model module includes a deep learning server deployed with a deep learning model M1, and the deep learning model module is used to generate a recommended journal list d2; The comparison module is used to compare the journal list d1 with the journal list d2, retain duplicate journals, delete non-duplicate journals, and obtain a recommended journal list d3; The journal database module is constructed with a journal database DB1, which contains journal data and published paper data; The crawler module is used to crawl the latest journal data from the Internet public database and website and save it in the journal database DB1; The journal verification module is used to compare the journal information in the journal list d3 with the journal data and published paper data in the journal database DB1 to complete the journal verification; at the same time, the latest journal information updated by the crawler module in the journal database DB1 is added to the journal list d3 to form a journal dictionary d4; The journal output module is used to output journal information, transmit the data of the journal dictionary d4 back to the user end and output it.

[0012] Compared with the prior art, the advantages of this application are: (1) This application discloses an intelligent journal recommendation system and method, which improves the prediction effect. Compared with the currently commonly used AI journal recommendation technology based on deep learning, this application combines traditional deep learning technology and more advanced large model technology. The information of the papers to be recommended is used to generate a journal list d1 through a large model. At the same time, the information of the papers to be recommended is used to generate a journal list d2 through a deep learning model M1. The journal list d1 and the journal list d2 are compared to obtain a journal list d3. Then, the journal list d3 is verified with the journal database DB1 to obtain a journal dictionary d4, and a journal recommendation scheme is obtained. This application avoids the problem that the use of a deep learning model alone has high requirements for data quality and data volume, and the accuracy of predicting recommended journals is not high, and effectively improves the prediction accuracy. (2) When constructing a deep learning model, this application concatenates journal text embedding, journal field feature embedding, paper text embedding, and paper field feature embedding to form a comprehensive feature vector; introduces an attention mechanism to assign different weights to different feature vectors to obtain fused features; processes the fused features through a series of fully connected layers to output the matching score of each journal; normalizes the output matching score, and obtains the probability distribution of each journal, i.e., the recommendation degree, based on the multi-label classification loss function, thereby avoiding the problem of too many categories and reducing the difficulty of prediction; (3) Since the present application method combines large model technology and crawler technology that support online search, it can timely update the latest journal information and paper publication information, avoiding prediction errors caused by untimely information updates; (4) This application compares the journal data in the journal database DB1 of the compared journal list d3 with the published paper data to complete the journal verification, which effectively solves the hallucination problem common in large models, avoids predicting wrong or non-existent journals, and improves the prediction accuracy; (5) This application builds ElasticSearch text search technology on the journal database DB1 to effectively improve the overall prediction speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 , a flowchart of an intelligent journal recommendation method of the present application; Figure 2 , a structural diagram of an intelligent journal recommendation system of the present application. DETAILED DESCRIPTION

[0014] The specific implementation of the present application is described below in conjunction with embodiments: It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions under which this application can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in this application without affecting the effects and purposes that can be achieved by this application.

[0015] like Figure 1 As shown, the present application discloses an intelligent journal recommendation method, comprising the following steps: Step 1: Build a journal database DB1, which contains journal data and published paper data; Step 2: Build a crawler module to crawl the latest journal data from the Internet public database and website and save it to the journal database DB1; Step 3: Based on the journal data and published paper data in the journal database DB1 and based on the pytorch framework, train the deep learning model M1; Step 4: Build the ElasticSearch search engine E1 on the journal database DB1; Step 5: The input module receives the information of the paper to be recommended input by the user; Step 6: The input module combines the information of the paper to be recommended with the journal prediction prompt words, and inputs them into the big model through the big model interface. The big model generates a recommended journal list d1 in the form of [journal 1, journal 2, journal 3]. The journals in d1 are sorted according to the journal recommendation degree given by the big model. Step 7: The input module inputs the information of the papers to be recommended into the deep learning model M1, and generates a recommended journal list d2 in the form of [journal 1, journal 2, journal 4]. The journals in d2 are sorted according to the journal recommendation degree given by the deep learning model M1; Step 8: Compare the journal list d1 with the journal list d2, keep the duplicate journals, delete the non-duplicate journals, and obtain the recommended journal list d3; that is, compare d1 [journal 1, journal 2, journal 3] with d2 [journal 1, journal 2, journal 4], keep the duplicate journals, delete the non-duplicate journals, and obtain the recommended journal list d3. The format of d3 is as follows: [journal 1, journal 2] (journal 1 and journal 2 of d1 and d2 are the same, so they are kept in d3 and the rest are deleted) Step 9: Based on the ElasticSearch retrieval engine E1 and the journal database DB1, the journal information in the journal list d3 is compared with the journal data and published paper data in the journal database DB1 to complete the journal verification, remove the false and non-existent journals in d3, and at the same time, the latest journal information (for example: journal 5) updated by the crawler module in the journal database DB1 is added to the journal list d3 to form a journal dictionary d4; the format of d4 is as follows: {journal 1: journal 1 data, journal 2: journal 2 data, journal 5: journal 5 data}.

[0016] Step 10: The data of the journal dictionary d4 is transmitted back to the user end of the journal recommendation system and output to obtain a journal recommendation scheme.

[0017] Preferably, the journal data in step 1 includes journal name, introduction, target readership, subject area, impact factor, publisher, publication cycle, A partition, B partition, publication ratio, review cycle, page charge, and whether it is Open Access; the published paper data includes paper title, abstract, full text, subject area, keywords, subject tags, target journal impact factor, and target partition.

[0018] Preferably, in step 2, the crawler module legally and compliantly crawls the latest journal data from public databases and websites on the Internet, saves it to the journal database DB1, and fully updates it once a week to ensure the real-time and accuracy of the data.

[0019] Preferably, the training process of the deep learning model M1 in step 3 includes the following steps: Step 3-1, data preparation: prepare the journal data and published paper data to be trained, and perform data preprocessing so that they can be fully and correctly input into the model; Step 3-2, model design: According to the journal data to be input and the format of published paper data, as well as the format of the journal data to be output, the structure of the deep learning model is designed, including a text embedding module, a domain feature embedding module, a feature fusion module, and an output classification module; The text embedding module includes journal text embedding and paper text embedding; Journal text embedding: Use a pre-trained language model to embed the journal name, introduction, and target readership to obtain the semantic representation of the journal; The pre-trained language models are BERT and RoBERTa.

[0020] Paper text embedding: Use the pre-trained language model to embed the paper title, abstract, and keywords to obtain the semantic representation of the paper; The domain feature embedding module includes journal domain feature embedding and paper domain feature embedding; Embedding journal domain features: embedding the journal’s subject field, impact factor, and review cycle to obtain the journal’s domain feature representation; Embedding the field features of the paper: embed the field, keywords, and topic tags of the paper to obtain the field feature representation of the paper; The feature fusion module includes splicing fusion and attention mechanism; Splicing and fusion: Splicing journal text embedding, journal field feature embedding, paper text embedding, and paper field feature embedding to form a comprehensive feature vector; Attention mechanism: Introduce the attention mechanism to assign different weights to different feature vectors and obtain the fused features; The output classification module includes a fully connected layer and a Softmax layer; Fully connected layer: The fused features are processed through a series of fully connected layers to output the matching score of each journal; Softmax layer: Normalizes the output matching scores and converts them into the predicted probability value of each journal, i.e., the recommendation degree. At the same time, the difference between the matching probability of the journal predicted by the model and the actual label is evaluated through the multi-label classification loss function. Step 3-3, model training and optimization: Input journal data and published paper data into the deep learning model for training. During the training process, use model evaluation indicators to evaluate the model performance; based on the evaluation results, adjust the model structure and training strategy, and repeatedly optimize the model parameters and hyperparameters until the model indicators reach the expected standards; in addition, use methods such as cross-validation to further verify the generalization ability of the model to ensure its reliability and effectiveness in practical applications.

[0021] The model evaluation indicators are accuracy, precision, recall and F1 score.

[0022] Step 3-4, model deployment: deploy the trained deep learning model to the deep learning server and open the interface so that the deep learning model can receive the information of the recommended papers in step 5 and output the predicted journal list d2.

[0023] Since a paper may be suitable for multiple journals, a multi-label classification loss function such as the binary cross entropy loss function is adopted.

[0024] Preferably, the multi-label classification loss function is defined as:

[0025] in: is the actual label, 0 or 1, indicating whether the paper is suitable for the i-th journal; is the matching probability of the journal predicted by the model.

[0026] This formula calculates the cross entropy loss for each journal and sums it over all journals. In other words, the loss function measures the difference between the probability value predicted by the model and the actual label. By minimizing the loss function, the model can gradually improve the accuracy of the prediction, making the matching degree of papers to multiple journals more accurate.

[0027] Preferably, in step 6, combining the information of the paper to be recommended with the journal prediction prompt word specifically comprises: inserting the title, field and abstract of the information of the paper to be recommended into the curly brackets {} after the prediction prompt word respectively.

[0028] Preferably, the large model in step 6 is the DeepSeek-R1-671B large model, and the recommendation degree refers to the matching degree between each journal in the journal list d1 generated by the DeepSeek-R1-671B large model and the paper to be recommended. The DeepSeek-R1-671B large model generates the recommendation degree while generating the journal list d1.

[0029] The recommendation degree refers to the matching degree between each journal and the paper in the journal list generated by the big model. The recommendation degree can be generated by the big model while generating the journal list d1, such as {journal 1: recommendation degree 96, journal 2: recommendation degree 87, journal 3: recommendation degree 77}.

[0030] like Figure 2 As shown, preferably, the present application discloses an intelligent journal recommendation system, including an input module, a large model module, a deep learning model module, a comparison module, a journal verification module, a journal database module, a crawler module and a journal output module; The input module is used for users to input information about papers to be recommended, combine the information about papers to be recommended with journal prediction prompt words, input the information about papers to be recommended into the big model module through the big model interface, and also input the information about papers to be recommended into the deep learning model module; The large model module is used to generate a recommended journal list d1; The deep learning model module includes a deep learning server deployed with a deep learning model M1, and the deep learning model module is used to generate a recommended journal list d2; The comparison module is used to compare the journal list d1 with the journal list d2, retain duplicate journals, delete non-duplicate journals, and obtain a recommended journal list d3; The journal database module is constructed with a journal database DB1, which contains journal data and published paper data; The crawler module is used to crawl the latest journal data from the Internet public database and website and save it in the journal database DB1; The journal verification module is used to compare the journal information in the journal list d3 with the journal data and published paper data in the journal database DB1 to complete the journal verification; at the same time, the latest journal information updated by the crawler module in the journal database DB1 is added to the journal list d3 to form a journal dictionary d4; The journal output module is used to output journal information, transmit the data of the journal dictionary d4 back to the user end and output it.

[0031] Example 1 There is a paper with the title {a1}, the field it belongs to {a2}, the keywords of the paper {a3}, and the abstract of the paper {a4}. Fill the data [a1, a2, a3, a4] into the input module (web page) of the journal recommendation system. Insert the paper information [a1, a2, a3, a4] into the journal prediction prompt words to get the wording w (You are a senior journal editor and you have a paper. The title of the paper is {a1}, the field of the paper is {a2}, the keywords of the paper are {a3}, and the abstract of the paper is {a4}. Based on the above paper information, please give a list of journals suitable for submission and sort the journals by matching degree). Input the wording w into the big model, and the big model outputs the result d1 {Based on the paper information you provided, the recommended journals are as follows: Journal 1, Journal 2, Journal 3, Journal 4, Journal 5}; Input the paper information [a1, a2, a3, a4] into the deep learning model M1, and the deep learning model M1 outputs the result d2 {journal 1, journal 3, journal 7, journal 6, journal 5}; Compare d1 and d2 and keep the duplicate journals, delete the non-duplicate journals, and get d3, d3 is {journal 1, journal 3, journal 5} Input d3 into the ElasticSearch text retrieval engine to search and verify the journal information in the journal database DB1. The search found that journal 5 in d3 does not exist in the journal database, so journal 5 is deleted from d3, and the detailed data of the journal (including the journal impact factor, the A partition, the publication cycle, etc.) is added to d3 to generate a journal dictionary d4, d4 is such as {journal 1: journal 1 data, journal 3: journal 3 data}; Send d4 back to the user.

[0032] Comparative Example The existing deep learning technology is used to predict the papers of the embodiment, and the results obtained are: The results obtained are: {Journal 6, Journal 8}.

[0033] Actual situation: The paper was actually successfully published in Journal 3, and the results were consistent with Example 1 but inconsistent with the comparative example, so the prediction accuracy of this application is high.

[0034] The present application discloses an intelligent journal recommendation system and method, which improves the prediction effect. Compared with the currently commonly used AI journal recommendation technology based on deep learning, the present application combines traditional deep learning technology and more advanced large model technology, generates a journal list d1 through a large model for the information of papers to be recommended, and generates a journal list d2 through a deep learning model M1 for the information of papers to be recommended. The journal list d1 and the journal list d2 are compared to obtain a journal list d3, and then the journal list d3 is verified to obtain a journal dictionary d4, thereby obtaining a journal recommendation scheme. The present application avoids the problem of high requirements on data quality and data volume and low accuracy in predicting recommended journals when simply using a deep learning model, and effectively improves the prediction accuracy.

[0035] When constructing a deep learning model, this application concatenates journal text embedding, journal field feature embedding, paper text embedding, and paper field feature embedding to form a comprehensive feature vector; and introduces an attention mechanism to assign different weights to different feature vectors to obtain fused features; the fused features are processed through a series of fully connected layers to output the matching score of each journal; the output matching score is normalized, and the probability distribution of each journal, i.e., the recommendation degree, is obtained according to the multi-label classification loss function, thereby avoiding the problem of too many categories and reducing the difficulty of prediction.

[0036] Since the application method combines large model technology and crawler technology that support online search, it can update the latest journal information and paper publication information in a timely manner, avoiding prediction errors caused by untimely information updates.

[0037] This application compares the journal data in the journal database DB1 of the compared journal list d3 with the published paper data to complete the journal verification, which effectively solves the illusion problem common in large models, avoids predicting the generation of erroneous or non-existent journals, and improves the prediction accuracy.

[0038] This application builds ElasticSearch text search technology on the journal database DB1 to effectively improve the overall prediction speed.

[0039] The preferred embodiments of the present application are described in detail above, but the present application is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present application.

[0040] Many other changes and modifications can be made without departing from the concept and scope of the present application.It should be understood that the present application is not limited to specific embodiments, and the scope of the present application is defined by the appended claims.

Claims

1. An intelligent journal recommendation method, characterized in that: The following steps are involved: Step 1: Build a journal database DB1, which contains journal data and published paper data; Step 2: Build a crawler module to crawl the latest journal data from the Internet public database and website and save it to the journal database DB1; Step 3: Based on the journal data and published paper data in the journal database DB1 and based on the pytorch framework, train the deep learning model M1; Step 4: Build the ElasticSearch search engine E1 on the journal database DB1; Step 5: The input module receives the information of the paper to be recommended input by the user; Step 6: The input module combines the information of the paper to be recommended with the journal prediction prompt words, and inputs them into the big model through the big model interface. The big model generates a recommended journal list d1, and the journals in d1 are sorted according to the journal recommendation degree given by the big model; Step 7: The input module inputs the information of the papers to be recommended into the deep learning model M1, and generates a recommended journal list d2. The journals in d2 are sorted according to the journal recommendation degree given by the deep learning model M1; Step 8: Compare the journal list d1 and the journal list d2, retain the duplicate journals, delete the non-duplicate journals, and obtain the recommended journal list d3; Step 9: Based on the ElasticSearch search engine E1 and the journal database DB1, the journal information in the journal list d3 is compared with the journal data and published paper data in the journal database DB1 to complete the journal verification. At the same time, the latest journal information updated by the crawler module in the journal database DB1 is added to the journal list d3 to form a journal dictionary d4. Step 10: The data of the journal dictionary d4 is transmitted back to the user end of the journal recommendation system and output to obtain a journal recommendation scheme.

2. The intelligent journal recommendation method according to claim 1, characterized in that: The journal data in step 1 include journal name, introduction, target readership, subject area, impact factor, publisher, publication cycle, A division, B division, publication ratio, review cycle, page charge, and whether it is Open Access; the published paper data include paper title, abstract, full text, subject area, keywords, subject tags, target journal impact factor, and target division.

3. The intelligent journal recommendation method according to claim 1, characterized in that: In step 2, the crawler module legally and compliantly crawls the latest journal data from the Internet public databases and websites, saves it to the journal database DB1, and updates it in full once a week.

4. The intelligent journal recommendation method according to claim 2, characterized in that: The training process of the deep learning model M1 in step 3 includes the following steps: Step 3-1, data preparation: prepare the journal data and published paper data to be trained, and perform data preprocessing so that they can be fully and correctly input into the model; Step 3-2, model design: According to the journal data to be input and the format of published paper data, as well as the format of the journal data to be output, the structure of the deep learning model is designed, including a text embedding module, a domain feature embedding module, a feature fusion module, and an output classification module; The text embedding module includes journal text embedding and paper text embedding; Journal text embedding: Use a pre-trained language model to embed the journal name, introduction, and target readership to obtain the semantic representation of the journal; Paper text embedding: Use the pre-trained language model to embed the paper title, abstract, and keywords to obtain the semantic representation of the paper; The domain feature embedding module includes journal domain feature embedding and paper domain feature embedding; Embedding journal domain features: embedding the journal’s subject field, impact factor, and review cycle to obtain the journal’s domain feature representation; Embedding the field features of the paper: embed the field, keywords, and topic tags of the paper to obtain the field feature representation of the paper; The feature fusion module includes splicing fusion and attention mechanism; Splicing and fusion: Splicing journal text embedding, journal field feature embedding, paper text embedding, and paper field feature embedding to form a comprehensive feature vector; Attention mechanism: Introduce the attention mechanism to assign different weights to different feature vectors and obtain the fused features; The output classification module includes a fully connected layer and a Softmax layer; Fully connected layer: The fused features are processed through a series of fully connected layers to output the matching score of each journal; Softmax layer: Normalizes the output matching scores and converts them into the predicted probability value of each journal, i.e., the recommendation degree. At the same time, the difference between the matching probability of the journal predicted by the model and the actual label is evaluated through the multi-label classification loss function. Step 3-3, model training and optimization: input journal data and published paper data into the deep learning model for training. During the training process, the model performance is evaluated using model evaluation indicators. Based on the evaluation results, the model structure and training strategy are adjusted, and the model parameters and hyperparameters are repeatedly optimized until the model indicators reach the expected standards. Step 3-4, model deployment: deploy the trained deep learning model to the deep learning server and open the interface so that the deep learning model can receive the information of the recommended papers in step 5 and output the predicted journal list d2.

5. The intelligent journal recommendation method according to claim 4, characterized in that: The multi-label classification loss function is defined as: ; in: is the actual label, 0 or 1, indicating whether the paper is suitable for the i-th journal; is the matching probability of the journal predicted by the model.

6. The intelligent journal recommendation method according to claim 1, characterized in that: In step 6, the paper information to be recommended is combined with the journal prediction prompt words as follows: the paper title, field and abstract of the paper information to be recommended are respectively inserted into the curly brackets {} after the prediction prompt words.

7. The intelligent journal recommendation method according to claim 1, characterized in that: The large model in step 6 is the DeepSeek-R1-671B large model, and the recommendation degree refers to the matching degree between each journal in the journal list d1 generated by the DeepSeek-R1-671B large model and the paper to be recommended. The DeepSeek-R1-671B large model generates the recommendation degree while generating the journal list d1.

8. An intelligent journal recommendation system, characterized by: It includes input module, large model module, deep learning model module, comparison module, journal verification module, journal database module, crawler module and journal output module; The input module is used for users to input information about papers to be recommended, combine the information about papers to be recommended with journal prediction prompt words, input the information about papers to be recommended into the big model module through the big model interface, and also input the information about papers to be recommended into the deep learning model module; The large model module is used to generate a recommended journal list d1; The deep learning model module includes a deep learning server deployed with a deep learning model M1, and the deep learning model module is used to generate a recommended journal list d2; The comparison module is used to compare the journal list d1 with the journal list d2, retain duplicate journals, delete non-duplicate journals, and obtain a recommended journal list d3; The journal database module is constructed with a journal database DB1, which contains journal data and published paper data; The crawler module is used to crawl the latest journal data from the Internet public database and website and save it in the journal database DB1; The journal verification module is used to compare the journal information in the journal list d3 with the journal data and published paper data in the journal database DB1 to complete the journal verification; at the same time, the latest journal information updated by the crawler module in the journal database DB1 is added to the journal list d3 to form a journal dictionary d4; The journal output module is used to output journal information, transmit the data of the journal dictionary d4 back to the user end and output it.

Images