Big data-based journal matching recommendation method and device

Abstract

The invention belongs to the technical field of computers, and discloses a journal matching recommendation method and device based on big data, and the method comprises the steps: obtaining a to-be-detected manuscript, and extracting feature information in the manuscript to form a target feature vector; acquiring historical periodical data, and extracting feature information of each periodical to form a corresponding periodical feature vector; training a machine learning model based on each journal feature vector to obtain a journal matching model; inputting the target feature vector into a journal matching model to obtain a plurality of matched journals and corresponding matching degrees; calculating the similarity between the target feature vector and the journal feature vector of each matched journal; and sorting the matched periodicals based on the similarity and the matching degree to obtain a recommended periodical list. According to the method, the influence of human subjective factors in model matching and the calculation amount required by similarity calculation can be reduced, and the accuracy and speed of journal matching are greatly improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a journal matching recommendation method and device based on big data. Background Art

[0002] In the current various journal statistical databases or paper management systems, intelligent matching of related journal data based on documents uploaded by users for researchers to consult and study is a common function. The current intelligent matching methods mainly include model matching based on various tags (such as keywords, etc.) pre-set on each journal, or matching methods based on text similarity.

[0003] However, matching relevant journals by traversing labels through a large model is heavily dependent on the accuracy of label settings. When the keywords set by the journal author are inappropriate, the journal labels will also be inaccurate, affecting the accuracy of the model output. The method of calculating similarity requires processing a large amount of journal data, resulting in long calculation time and high computing power consumption. Summary of the invention

[0004] The present application provides a journal matching recommendation method and device based on big data, which can reduce the influence of human subjective factors in model matching and greatly improve the accuracy and speed of journal matching.

[0005] In a first aspect, the present application embodiment provides a journal matching recommendation method based on big data, comprising:

[0006] Obtain the document to be detected, extract the feature information therein to form a target feature vector;

[0007] Obtain historical journal data, extract characteristic information of each journal to form a corresponding journal characteristic vector;

[0008] Train the machine learning model based on the feature vectors of each journal to obtain the journal matching model;

[0009] Input the target feature vector into the journal matching model to obtain multiple matching journals and corresponding matching degrees;

[0010] Calculate the similarity between the target feature vector and the journal feature vector of each matching journal;

[0011] The matching journals are sorted based on similarity and matching degree to obtain a list of recommended journals.

[0012] Furthermore, the method also includes:

[0013] Before extracting the feature information of the document to be detected, removing the preset stop words in the document to be detected;

[0014] Perform word segmentation, stem restoration and lowercase conversion on the document to be tested.

[0015] Furthermore, the above-mentioned obtaining the document to be detected and extracting the feature information therein to form a target feature vector includes:

[0016] Extract the emotional feature vector of the document to be detected based on the preset emotional dictionary;

[0017] The TF-IDF method is used to extract the word frequency feature vector of the document to be detected;

[0018] Combine the sentiment feature vector and the word frequency feature vector to obtain the target feature vector.

[0019] Furthermore, the above-mentioned extraction of emotional features of the document to be detected based on the preset emotional dictionary includes:

[0020] Assign a sentiment polarity score to each word in the document to be tested according to the preset sentiment dictionary;

[0021] The sentiment score of the document to be detected is calculated based on each sentiment polarity score;

[0022] Determine the emotion type and emotion intensity of the document to be tested according to the document emotion score;

[0023] The document sentiment score, sentiment type and sentiment intensity are constructed as a sentiment feature vector.

[0024] Furthermore, the method also includes:

[0025] Before training the machine learning model, the chi-square statistic of each journal feature vector was calculated;

[0026] Construct a contingency table corresponding to each journal feature vector; the rows of the contingency table represent the values ​​of different features in the journal feature vector, and the columns represent the categories of different features;

[0027] Determine the chi-square threshold of the journal feature vector based on the preset significance level and contingency table;

[0028] If the chi-square statistic is less than the chi-square threshold, the corresponding journal feature vector is eliminated.

[0029] Furthermore, the chi-square threshold of the journal feature vector is determined according to the preset significance level and the contingency table, including:

[0030] Subtract 1 from the number of rows and columns in the contingency table, and then multiply them to get the degrees of freedom;

[0031] Multiply the preset significance level and the degrees of freedom to obtain the chi-square threshold.

[0032] Furthermore, the above calculation of the similarity between the target feature vector and the journal feature vectors of each matching journal includes:

[0033] The Mankadon distance or Jaccard similarity coefficient between the target feature vector and the journal feature vector is calculated, and the Mankadon distance or Jaccard similarity coefficient is used as the similarity.

[0034] Furthermore, the matching journals are sorted based on similarity and matching to obtain a list of recommended journals, including:

[0035] Normalize each similarity and each matching degree respectively;

[0036] Add the normalized similarity and matching degree to get the ranking value of the matching journals;

[0037] Sort each matching journal according to the ranking value to obtain a recommended journal list.

[0038] Furthermore, the method also includes:

[0039] The popularity score of each matching journal is calculated based on the knowledge graph method to obtain a popularity ranking list;

[0040] Adjust the ranking of each matching journal in the recommended journal list according to the popularity ranking list.

[0041] Furthermore, the above-mentioned knowledge graph-based method calculates the popularity score of each matching journal and obtains a popularity ranking list, including:

[0042] Each matching journal is taken as a node, the journal feature vector is taken as the attribute of the corresponding node, and the citation relationship between each matching journal is taken as the edge connecting each node to obtain the journal knowledge graph;

[0043] Get the comment popularity score based on the comment text and corresponding comment time of each node;

[0044] The collection popularity score is obtained based on the accumulated collection volume and collection growth rate of each node;

[0045] Perform weighted calculation on the comment heat score and the collection heat score to get the node score;

[0046] A graph algorithm is used to calculate the heat score of each node based on the scores of each node and the edges between the nodes;

[0047] Sort the matching journals corresponding to each node according to the popularity score to obtain a popularity ranking list.

[0048] Furthermore, the collection popularity score is obtained according to the cumulative collection volume and collection growth rate of each node, including:

[0049] Determine a first score of the collection interval where the cumulative collection amount is located, and a second score of the growth rate interval where the collection growth rate is located;

[0050] Add the first score and the second score to get the collection popularity score.

[0051] Furthermore, the above-mentioned adjusting the ranking of each matching journal in the recommended journal list according to the popularity ranking list includes:

[0052] The average ranking of the matching journals in the popularity ranking list and the recommended journal list is taken as the new ranking of the matching journals in the recommended journal list.

[0053] Furthermore, the method also includes:

[0054] Get the number of clicks for each matching journal, and modify the corresponding matching degree based on the number of clicks;

[0055] The journal matching model is trained based on the modified matching degree and the corresponding journal feature vector.

[0056] In a second aspect, the embodiment of the present application provides a journal matching recommendation device based on big data, comprising:

[0057] The target document module is used to obtain the document to be detected and extract the feature information therein to form a target feature vector;

[0058] The journal module is used to obtain historical journal data and extract the characteristic information of each journal to form the corresponding journal feature vector;

[0059] A training module is used to train a machine learning model based on the feature vectors of each journal to obtain a journal matching model;

[0060] A matching module, used for inputting the target feature vector into the journal matching model to obtain multiple matching journals and corresponding matching degrees;

[0061] A calculation module, used for calculating the similarity between the target feature vector and the journal feature vectors of each matching journal;

[0062] The recommendation module is used to sort the matching journals based on similarity and matching degree to obtain a list of recommended journals.

[0063] Furthermore, the device also includes a preprocessing module, which is used to remove preset stop words in the document to be detected before extracting the feature information of the document to be detected; and to perform word segmentation, stem restoration and lowercase conversion on the document to be detected.

[0064] Furthermore, the target document module includes:

[0065] The sentiment unit is used to extract the sentiment feature vector of the document to be detected based on a preset sentiment dictionary;

[0066] The word frequency unit is used to extract the word frequency feature vector of the document to be detected by using the TF-IDF method;

[0067] The combining unit is used to combine the sentiment feature vector and the word frequency feature vector to obtain the target feature vector.

[0068] Furthermore, the above emotion unit is used for:

[0069] Assign a sentiment polarity score to each word in the document to be tested according to the preset sentiment dictionary;

[0070] The sentiment score of the document to be detected is calculated based on each sentiment polarity score;

[0071] Determine the emotion type and emotion intensity of the document to be tested according to the document emotion score;

[0072] The document sentiment score, sentiment type and sentiment intensity are constructed as a sentiment feature vector.

[0073] Furthermore, the device also includes:

[0074] A statistics module, used to calculate the chi-square statistic of each journal feature vector before training the machine learning model;

[0075] The contingency table module is used to construct the contingency table corresponding to each journal’s feature vector;

[0076] Threshold module, used to determine the chi-square threshold of journal feature vector according to the preset significance level and contingency table;

[0077] The elimination module is used to eliminate the corresponding journal feature vector if the chi-square statistic is less than the chi-square threshold.

[0078] Furthermore, the threshold module includes:

[0079] The degree of freedom unit is used to reduce the number of rows and columns of the contingency table by 1 respectively, and then multiply them to get the degree of freedom;

[0080] The threshold unit is used to multiply the preset significance level and the degrees of freedom to obtain the chi-square threshold.

[0081] Furthermore, the above-mentioned calculation module is used to calculate the Mankard distance or Jaccard similarity coefficient between the target feature vector and the journal feature vector, and use the Mankard distance or Jaccard similarity coefficient as the similarity.

[0082] Furthermore, the above recommendation module includes:

[0083] A normalization unit, used for normalizing each similarity and each matching degree respectively;

[0084] The ranking value unit is used to add the normalized similarity and matching degree to obtain the ranking value of the matching journal;

[0085] The recommendation list unit is used to sort each matching journal according to the sorting value to obtain a recommended journal list.

[0086] Furthermore, the device also includes:

[0087] The popularity ranking module is used to calculate the popularity score of each matching journal based on the knowledge graph method and obtain a popularity ranking list;

[0088] The sorting update module is used to adjust the sorting of each matching journal in the recommended journal list according to the popularity sorting list.

[0089] Furthermore, the heat ranking module includes:

[0090] A graph construction unit is used to take each matching journal as a node, take the journal feature vector as the attribute of the corresponding node, take the citation relationship between each matching journal as the edge connecting each node, and obtain a journal knowledge graph;

[0091] The comment heat unit is used to obtain the comment heat score according to the comment text and corresponding comment time of each node;

[0092] The collection heat unit is used to obtain the collection heat score according to the accumulated collection volume and collection growth rate of each node;

[0093] The node score unit is used to perform weighted calculation on the comment heat score and the collection heat score to obtain the node score;

[0094] A node heat unit is used to calculate the heat score of each node based on the scores of each node and the edges between the nodes using a graph algorithm;

[0095] The heat ranking unit is used to sort the matching journals corresponding to each node according to the heat score to obtain a heat ranking list.

[0096] Furthermore, the collection heat unit is used to:

[0097] Determine a first score of the collection interval where the cumulative collection amount is located, and a second score of the growth rate interval where the collection growth rate is located;

[0098] Add the first score and the second score to get the collection popularity score.

[0099] Furthermore, the ranking update module is used to obtain the average ranking of the matching journals in the popularity ranking list and the recommended journal list as a new ranking of the matching journals in the recommended journal list.

[0100] Furthermore, the device also includes:

[0101] Click acquisition module, used to obtain the number of clicks of each matching journal and modify the corresponding matching degree based on the number of clicks;

[0102] A feedback training module is used to train the journal matching model based on the modified matching degree and the corresponding journal feature vector.

[0103] In a third aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a journal matching recommendation method based on big data as described in any of the above embodiments are performed.

[0104] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a journal matching recommendation method based on big data as in any of the above embodiments.

[0105] In summary, compared with the prior art, the technical solution provided in the embodiment of the present application has at least the following beneficial effects:

[0106] The embodiment of the present application provides a journal matching recommendation method based on big data. First, the journal matching model is trained through the journal feature vector to obtain multiple matching journals and corresponding matching degrees. Then, the similarity between the target feature vector of the manuscript to be detected and the journal feature vector of each matching journal is calculated. Finally, the similarity and matching degree are combined to form a recommended journal list. The above method combines the similarity and the matching method of the large model based on the feature vector, which not only avoids the situation where the model output is affected by the inaccurate labels set by humans, greatly improves the accuracy of journal matching, but also reduces the amount of data required for similarity calculation, that is, the manuscript to be detected only needs to perform similarity calculation with the matching journal output by the model, thereby improving the matching speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 A flowchart of a journal matching recommendation method based on big data is provided as an exemplary embodiment of the present application.

[0108] Figure 2 A flowchart of target feature vector construction steps provided for an exemplary embodiment of the present application.

[0109] Figure 3 A flowchart of the emotion feature vector steps provided for an exemplary embodiment of the present application.

[0110] Figure 4 A flowchart of the chi-square verification steps provided for an exemplary embodiment of the present application.

[0111] Figure 5 A flowchart of the journal sorting steps provided for an exemplary embodiment of the present application.

[0112] Figure 6 A flowchart of the heat sorting steps provided for an exemplary embodiment of the present application.

[0113] Figure 7 A flowchart of the collection popularity calculation steps provided for an exemplary embodiment of the present application.

[0114] Figure 8 A flowchart of the model iteration steps provided for an exemplary embodiment of the present application.

[0115] Fig. 9 A structural diagram of a journal matching recommendation device based on big data provided as an exemplary embodiment of the present application.

[0116] Fig.10 A structural diagram of a target document module provided for an exemplary embodiment of the present application.

[0117] Fig.11 A structural diagram of a recommendation module provided for an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0118] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0119] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0120] Terminology explanation:

[0121] 1) Similarity: It is an indicator that measures the closeness of two vectors in the feature space. Similarity is usually used to indicate the similarity or difference between vectors. It has a wide range of applications in machine learning and data analysis, such as cluster analysis, recommendation systems, image recognition, etc. Similarity measurement methods include:

[0122] Cosine Similarity measures the similarity between two vectors by calculating the cosine of the angle between them. Cosine similarity focuses on the direction of the vector rather than its size.

[0123] Euclidean distance measures the straight-line distance between two vectors. In contrast to cosine similarity, Euclidean distance focuses on the size and direction of the vector. The smaller the Euclidean distance, the more similar the two feature vectors are.

[0124] Manhattan distance, also known as L1 distance, is the sum of the absolute values ​​of the differences in each dimension.

[0125] Jaccard Similarity is used to measure the similarity between two sets and is calculated by the size of their intersection and union.

[0126] Pearson Correlation Coefficient measures the linear correlation between two vectors.

[0127] Spearman's Rank Correlation Coefficient is a rank-based nonparametric method used to measure the monotonic association between two vectors.

[0128] Mahalanobis distance takes into account the spatial distribution of data and uses the covariance matrix to calculate the distance.

[0129] Levenshtein Distance measures the difference between two sequences by calculating the minimum number of single-character edits (insertions, deletions, or substitutions) required to transform one sequence into the other.

[0130] 2) Sentiment polarity score: refers to the quantitative representation of the positive or negative tendency of a word in sentiment analysis. This score is usually derived from a sentiment dictionary or predicted by a large model, and is used to determine the sentiment tendency expressed by each word in the text. The sentiment dictionary contains a large number of words and their corresponding sentiment polarity scores.

[0131] For example, AFINN, VADER (particularly used in the field of NLP for social media texts), and SentiWordNet are all commonly used sentiment lexicons. Sentiment polarity is usually divided into positive, negative, and neutral. Polarity scores usually range from -1 (most negative) to 1 (most positive), with 0 usually indicating neutral.

[0132] For a text, the sentiment of the entire text can be calculated by summing or averaging the sentiment polarity scores of each word in the text. The sentiment polarity of a word may depend on the context. For example, "bank" usually has a positive meaning, but in some cases (such as "bank bankruptcy"), it may have a negative meaning.

[0133] 3) Knowledge Graph: It is a structured semantic knowledge base that represents entities and their relationships in the form of a graph. The main features of the knowledge graph include:

[0134] Entities: Entities are basic concepts in the knowledge graph, which can be people, places, etc.

[0135] Relationships: Relationships define semantic connections between entities, such as "belongs to", "is located in", etc.

[0136] Attributes: Entities can have attributes that provide additional descriptive information about the entity, such as a person's date of birth, a country's language, etc.

[0137] Graph Structure: The knowledge graph is in the form of a graph, where nodes represent entities and edges represent relationships, forming a complex network structure.

[0138] Multi-source Integration: The data of the knowledge graph can come from multiple different information sources, including structured data, semi-structured data, and unstructured data.

[0139] Dynamic Updating: The knowledge graph can be continuously updated and improved based on newly acquired data.

[0140] Scalability: Knowledge graphs can be expanded to contain millions or even billions of entities and relationships.

[0141] 4) Graph algorithms: refers to algorithms that run on graph structured data and are used to perform various tasks, such as searching, analyzing, and mining information in knowledge graphs. The graph algorithms that calculate node importance mainly include the following:

[0142] Degree Centrality: Degree centrality is a measure of a node's importance by measuring the number of edges it is connected to. That is, the node with the most connected nodes is usually considered more important. In a directed graph, degree centrality is divided into in-degree and out-degree.

[0143] Closeness Centrality: Closeness centrality measures the average distance from a node to any other node. The closer a node is to all other nodes, the higher its closeness centrality. This can reflect the reachability of a node in the network.

[0144] Betweenness Centrality: Betweenness Centrality measures the number of times a node appears in all the shortest paths in the graph, that is, the number of times a node acts as a bridge for the shortest path between two other nodes. The higher the value, the more important the role the node plays in the information flow.

[0145] Eigenvector Centrality: Eigenvector centrality considers the importance of a node to be determined by the importance of its neighboring nodes. It is calculated iteratively until the eigenvector converges to a stable solution. This centrality takes into account the influence of the node and the degree to which it is connected by important nodes.

[0146] PageRank: The PageRank algorithm is based on an assumption that the importance of a node is determined by the number and quality of nodes pointing to it. It uses an iterative method to calculate the PageRank value of each node until convergence.

[0147] K-Core: K-Core is used to identify highly interconnected subgraphs in a graph, i.e. k-cores. In a k-core, each node is connected to at least k other nodes. K-Core can identify the core structure in a network.

[0148] HITS algorithm: This algorithm distinguishes two types of nodes: Hubs and Authorities. Hubs are nodes that point to multiple Authorities, while Authorities are nodes that are pointed to by multiple Hubs. The HITS algorithm iteratively calculates the Hubs and Authorities scores of each node.

[0149] See also Figure 1 , the embodiment of the present application provides a journal matching recommendation method based on big data, including:

[0150] Step S1, obtaining a document to be detected, and extracting feature information therein to form a target feature vector.

[0151] The documents to be tested are received documents that need to be matched with relevant journals. The content can be in any form and is not strictly required to be in the format of a paper or journal. It only needs to have text.

[0152] Step S2, obtaining historical journal data, extracting characteristic information of each journal to form a corresponding journal characteristic vector.

[0153] Specifically, because the journals published in the journal database are already in a standardized fixed format, when extracting journal feature information, the journal's technical field, impact factor, etc. can be directly used as feature information to construct the journal feature vector, or the same feature extraction and construction method as the above-mentioned target feature vector can be used.

[0154] Step S3, training a machine learning model based on the feature vectors of each journal to obtain a journal matching model.

[0155] Among them, the machine learning model can adopt models such as decision tree, support vector machine, K nearest neighbor, clustering algorithm, etc.

[0156] Specifically, in practical applications, due to different training methods, different machine learning models perform differently in classification and clustering. For example, decision trees infer target values ​​from data features by learning simple decision rules, so the training method is to build a tree structure, usually using a greedy algorithm such as ID3, C4.5 or CART; support vector machines find the optimal boundary between data points, and the training method is to use the Lagrange multiplier method and dual problem solving; K nearest neighbors is instance-based learning, which predicts by finding the K nearest neighbors of the test data point, and the training method is to store the training data and perform distance calculation and search during prediction; clustering algorithms such as K-Means and hierarchical clustering group data points into different clusters; the training method is iterative optimization, such as K-Means iteratively selecting cluster centers and redistributing data points. Therefore, multiple machine learning models can be trained separately using journal feature vectors, and the best performing model can be selected as the journal matching model based on the training results.

[0157] During the training process, the journal feature vector is input into the machine learning model, and the output data is the matching journal and the matching degree of each matching journal. Each matching journal is a journal associated with the journal corresponding to the input journal feature vector. The matching degree is the degree of association between the matching journal and the journal corresponding to the input journal feature vector. Therefore, the training goal is that the matching journal with the highest matching degree output by the model is the journal corresponding to the input journal feature vector. At this point, it can be regarded as the model has met expectations.

[0158] Step S4, inputting the target feature vector into the journal matching model to obtain multiple matching journals and corresponding matching degrees.

[0159] Specifically, the number of output matching journals is usually determined by a preset matching degree threshold within the journal matching model, that is, when the matching degree is lower than the preset matching degree threshold, the journal and the document to be detected are considered irrelevant.

[0160] Furthermore, a preset number of matching journals can be set within the journal matching model, and the matching journals can be sorted by the degree of matching, and the matching journals with the preset number of matching journals can be selected for output; however, although this method can further reduce the amount of similarity calculation in the next step, i.e., step S5, when there are many researchers in the technical field and a large number of journals, useful journal data may be missed, and therefore the utilization rate is low.

[0161] Step S5, calculating the similarity between the target feature vector and the journal feature vectors of each matching journal.

[0162] Among them, the method of calculating the similarity between two feature vectors may include: 1) cosine similarity, which measures the similarity of two vectors by calculating the cosine value of the angle between them; 2) Euclidean distance, which calculates the straight-line distance between two vectors in Euclidean space; 3) Jaccard similarity coefficient, which is used to measure the similarity of two sets and calculates the ratio of intersection and union; 4) Pearson correlation coefficient, which measures the linear correlation between two variables; 5) Mahalanobis distance, which depends on the inverse of the covariance matrix and is an effective distance measurement method; 6) KL divergence, which is used to measure the dissimilarity between two probability distributions.

[0163] Specifically, in step S5 of the present application, the object of similarity calculation is only the matching journals output by the journal matching model, rather than all the historical journals in the database. Therefore, compared with the prior art, the amount of calculation is greatly reduced.

[0164] Step S6, sorting the matching journals based on the similarity and matching degree to obtain a recommended journal list.

[0165] Specifically, the similarity is calculated directly using two feature vectors, while the matching degree is output by the model. This application combines the parameters obtained by the two methods to sort the matching journals, which will make the sorting results more accurate and reduce the deviation in the matching results caused by inaccurate calculation of any one of the parameters.

[0166] The above embodiment provides a journal matching recommendation method based on big data. First, the journal matching model is trained through the journal feature vector to obtain multiple matching journals and corresponding matching degrees. Then, the similarity between the target feature vector of the document to be detected and the journal feature vector of each matching journal is calculated. Finally, the similarity and matching degree are combined to form a recommended journal list. The above method combines the similarity and the matching method of the large model based on the feature vector, which not only avoids the situation where the model output is affected by the inaccurate labels set by humans, greatly improves the accuracy of journal matching, but also reduces the amount of data required for similarity calculation, that is, the document to be detected only needs to be calculated for similarity with the matching journal output by the model, thereby improving the matching speed.

[0167] In some embodiments, the method further includes:

[0168] Step S01, before extracting the feature information of the document to be detected, removing the preset stop words in the document to be detected.

[0169] Among them, the preset stop words refer to words that frequently appear in the text but contribute little to the meaning of the text, such as "de", "he", "shi", etc. Removing these words can reduce the noise when extracting feature information subsequently.

[0170] Step S02, performing word segmentation, stemming, and lowercase conversion on the document to be detected.

[0171] Among them, word segmentation refers to splitting the text in the document to be detected into individual words or phrases; stemming refers to converting words into their dictionary forms, for example, converting "better" to the comparative form of "good"; lowercase conversion is to convert all text to lowercase to eliminate the morphological differences caused by case.

[0172] In addition to the above-mentioned processing methods, operations such as removing phrases and synonyms, annotating the parts of speech of nouns, verbs, adjectives, etc., and removing punctuation marks and special symbols can also be added.

[0173] In the specific implementation process, the content of the document to be detected may contain a lot of information that is useless for journal matching or information that affects the matching accuracy. Therefore, before extracting the feature information in this application, performing the above-mentioned processing operations on the document to be detected can make the extraction of feature information faster and more accurate, thereby making the matching result more accurate.

[0174] Specifically, the word segmentation operation helps to identify the basic structure of the text, removing stop words can reduce data noise, improve the quality of features and the accuracy of analysis, stemming can retain the part-of-speech information of words, which helps to understand the meaning of the document to be detected, lowercase conversion can eliminate case differences, ensure the consistency of text processing, and avoid treating the same word as different words due to different cases, and removing special characters and punctuation can reduce the interference of irrelevant information and simplify the text data.

[0175] Please refer to Figure 2 , in some embodiments, obtaining the document to be detected and extracting the feature information therein to form a target feature vector may specifically include the following steps:

[0176] Step S11, extracting the sentiment feature vector of the document to be detected based on a preset sentiment dictionary.

[0177] Among them, the sentiment feature vector is used to represent the understanding of the emotional tendency in the document to be detected, which helps the journal matching model to more accurately capture the views and attitudes of the author of the document to be detected.

[0178] Specifically, in a certain technical field, there may be different views and opinions on a certain research direction; for example, for the application of a certain material in a certain product, Journal A may focus on the positive effects or advantages of the material in the product, while Journal B focuses on the negative effects or disadvantages of the material in the product.

[0179] Therefore, in the absence of sentiment features, important semantic information may be lost during the matching process, affecting the model's comprehensive understanding of the document to be detected, thereby reducing the accuracy of the matching.

[0180] Step S12: using the TF-IDF method to extract the word frequency feature vector of the document to be detected.

[0181] Specifically, all unique words are extracted from the document to be detected, and a vocabulary is established. The number of occurrences of each word in the vocabulary, that is, the word frequency, is calculated; for each word in the vocabulary, its inverse document frequency in the document to be detected is calculated.

[0182] For each word in each document, calculate its TF-IDF value, which is the product of the word frequency and the inverse document frequency, to obtain a word frequency feature vector, where each dimension in the word frequency feature vector corresponds to a word in the vocabulary.

[0183] Step S13, combining the sentiment feature vector and the word frequency feature vector to obtain a target feature vector.

[0184] The above embodiment adds a sentiment feature vector when constructing a target feature vector, and combines it with a word frequency feature vector, so that the journal matching model can provide more accurate matching and recommendation by analyzing the sentiment tendency of the manuscript to be detected.

[0185] See also Figure 3 In some embodiments, the above-mentioned extraction of emotional features of the document to be detected based on a preset emotional dictionary includes:

[0186] Step S111, assigning a sentiment polarity score to each word in the document to be detected according to a preset sentiment dictionary.

[0187] Existing sentiment dictionaries, such as AFINN, VADER, etc., can be used here to assign sentiment polarity scores to each word in the document to be tested based on the sentiment scores of different words in the dictionary.

[0188] Specifically, when the polarity is negative, the corresponding sentiment polarity score is a negative number, when the polarity is positive, the corresponding sentiment polarity score is a positive number, and when the polarity is neutral, the corresponding sentiment polarity score is 0.

[0189] Step S112: Calculate the document sentiment score of the document to be detected based on each sentiment polarity score.

[0190] Specifically, the sentiment polarity scores are weighted according to the frequency of each word, or the sentiment polarity scores are directly averaged to obtain the document sentiment score.

[0191] Step S113, determining the emotion type and emotion intensity of the document to be detected according to the document emotion score.

[0192] Among them, the emotion type is divided into positive, negative or neutral categories, which is usually obtained according to the numerical range of the document emotion score. For example, when the emotion score is less than the preset negative threshold, it is a negative emotion, when it is greater than a preset positive threshold, it is a positive emotion, and when it is between the preset negative threshold and the preset positive threshold, it is a neutral emotion. The emotion intensity is the difference between the document emotion score and the corresponding threshold. For negative emotion, it is the difference with the preset negative threshold, for positive emotion, it is the difference with the preset positive threshold, and for neutral emotion, it is the difference with 0. The difference is used as the emotion intensity.

[0193] Step S114, constructing the document sentiment score, sentiment type and sentiment intensity into a sentiment feature vector.

[0194] The above embodiment constructs a sentiment feature vector from the sentiment score, sentiment type and sentiment intensity of the manuscript, which ensures the comprehensiveness and accuracy of the sentiment analysis in the manuscript to be detected and helps the journal matching model to further perform accurate matching.

[0195] See also Figure 4 In some embodiments, the method further comprises:

[0196] Step S301, before training the machine learning model, calculate the chi-square statistic of each journal feature vector.

[0197] Specifically, the chi-square statistic x 2 The calculation formula is:

[0198]

[0199] Among them, O ij is the frequency of the word with coordinate (i, j) in the journal feature vector, E ij is the expected frequency under the corresponding independence assumption.

[0200] Step S302, constructing a contingency table corresponding to each journal feature vector; wherein the rows of the contingency table represent the values ​​of different features in the journal feature vector, and the columns represent the categories of different features.

[0201] Specifically, a contingency table is usually a two-dimensional table, and the cells of the table contain the observed frequencies or counts of corresponding category combinations. For example, suppose there are two categorical variables in the journal feature vector: gender (male and female) and smoking habits (smokers and non-smokers). A simple contingency table may look like this:

[0202]

[0203] In this table, a represents the number of male smokers, b represents the number of male non-smokers, c represents the number of female smokers, and d represents the number of female non-smokers. Through this contingency table, a chi-square test can be used to test whether there is a statistical association between gender and smoking habits. Therefore, this application uses a chi-square test to determine whether there is a correlation between the various feature information in the journal feature vector, so as to determine the quality of the journal corresponding to the journal feature vector.

[0204] Step S303, determining the chi-square threshold of the journal feature vector according to the preset significance level and the contingency table.

[0205] Among them, the preset significance level can usually be 0.05, which is used to determine whether the results of the chi-square test are significant.

[0206] Step S304: if the chi-square statistic is less than the chi-square threshold, the corresponding journal feature vector is eliminated.

[0207] In practical applications, due to the uneven quality of journals in the journal database, the quality of the corresponding generated journal feature vectors is also uneven. Therefore, in order to improve the training speed of the model and avoid some journal feature vectors with poor quality and poor effectiveness from not being able to produce good training effects and delaying training time, this application adopts the chi-square test method to screen out journal feature vectors with better effectiveness and ensure the training efficiency of the machine learning model.

[0208] In some embodiments, the above-mentioned determination of the chi-square threshold of the journal feature vector according to the preset significance level and the contingency table includes:

[0209] Step S3031, subtract 1 from the number of rows and columns of the contingency table, and then multiply them to obtain the degree of freedom.

[0210] Step S3032, multiplying the preset significance level and the degrees of freedom to obtain a chi-square threshold.

[0211] Specifically, the chi-square threshold = preset significance level × (number of rows - 1) × (number of columns - 1).

[0212] The chi-square test is a fast and effective method, especially when dealing with high-dimensional data sets. It can significantly reduce the number of invalid or low-quality features, thereby speeding up the training efficiency of the model.

[0213] In some embodiments, the above-mentioned calculation of the similarity between the target feature vector and the journal feature vectors of each matching journal includes:

[0214] The Mankadon distance or Jaccard similarity coefficient between the target feature vector and the journal feature vector is calculated, and the Mankadon distance or Jaccard similarity coefficient is used as the similarity.

[0215] Among them, Manhattan distance is also called L1 distance, which is the sum of the absolute values ​​of the differences in each dimension. The Jaccard similarity coefficient is calculated by the absolute value of the ratio of the intersection to the union of two eigenvectors.

[0216] Specifically, among the many similarity calculation methods, this application focuses on the calculation method of the Mankard distance or the calculation method of the Jaccard similarity coefficient. This is because the Mankard distance and the Jaccard similarity coefficient are fast to calculate, are robust to outliers, and are suitable for large-scale data sets, especially when the feature dimension is very high; therefore, the use of the above two methods will help to further improve the calculation speed of the similarity and further improve the speed of journal recommendation.

[0217] See also Figure 5 In some embodiments, the above-mentioned sorting of the matching journals based on similarity and matching degree to obtain a recommended journal list may specifically include the following steps:

[0218] Step S61, normalizing each similarity and each matching degree respectively.

[0219] Step S62, adding the normalized similarity and matching degree to obtain the ranking value of the matching journals.

[0220] Step S63, sorting each matching journal according to the sorting value to obtain a recommended journal list.

[0221] Specifically, since the similarity is calculated directly using two feature vectors, and the matching degree is output by the journal matching model, the matching degree formats output by different journal matching models may be different. Therefore, this application normalizes the two parameters separately before combining them to ensure the accuracy of the ranking value and the accuracy of the recommended journal list.

[0222] In some embodiments, the method further comprises:

[0223] Step S71, calculate the popularity score of each matching journal based on the knowledge graph method to obtain a popularity ranking list.

[0224] Step S72: adjusting the ranking of each matching journal in the recommended journal list according to the popularity ranking list.

[0225] Specifically, when recommending journals based on their similarity and matching, it is easy to mistakenly rank some older basic authoritative journals in the corresponding technical field in a relatively high position; in terms of academic research, the latest progress in a research direction in a certain technical field is often the focus of researchers in the corresponding research direction.

[0226] Therefore, the present application further takes the popularity of the journal into consideration when ranking the journal recommendation list, so that when recommending journals with a higher degree of relevance, the more popular research content in the corresponding technical field can be recommended at the front.

[0227] See also Figure 6 In some embodiments, the above-mentioned knowledge graph-based method calculates the popularity score of each matching journal to obtain a popularity ranking list, which may specifically include the following steps:

[0228] Step S711, taking each matching journal as a node, taking the journal feature vector as the attribute of the corresponding node, taking the citation relationship between each matching journal as the edge connecting each node, and obtaining a journal knowledge graph.

[0229] Specifically, considering that each matching journal most likely belongs to the same technical field or even the same research direction, there is a high possibility that there is a mutual citation relationship between them. Therefore, this application uses a knowledge graph to associate matching journals with citation relationships, thereby ensuring the accurate calculation of the heat scores of matching journals.

[0230] Step S712, obtaining the comment popularity score according to the comment text of each node and the corresponding comment time.

[0231] Specifically, a comment curve is constructed based on the number of new comment texts under the journal corresponding to the node and the corresponding comment time within the same day, and the comment curve is derived to obtain the comment heat score.

[0232] Step S713, obtaining a collection heat score according to the accumulated collection amount and collection growth rate of each node.

[0233] Step S714, perform weighted calculation on the comment heat score and the collection heat score to obtain the node score.

[0234] Step S715: Calculate the heat score of each node based on the score of each node and the edges between the nodes using a graph algorithm.

[0235] Specifically, graph algorithms (such as PageRank, HITS, Betweenness Centrality, etc.) can be used to calculate the importance of each node. This application equates "importance" with the popularity score of the matching journal.

[0236] Step S716, sort the matching journals corresponding to each node according to the popularity score to obtain a popularity ranking list.

[0237] The above embodiment determines the node score of each node based on the node, that is, the comment status and collection status of the matching journal, and further calculates the heat score of the node, which can ensure the calculation accuracy and consistency of the heat score.

[0238] See also Figure 7 In some embodiments, obtaining the collection popularity score according to the accumulated collection amount and collection growth rate of each node may specifically include the following steps:

[0239] Step S7141, determine the first score of the collection interval where the cumulative collection amount is located, and the second score of the growth rate interval where the collection growth rate is located. The collection growth rate is the number of collections added in the preset time period before the current moment divided by the preset time period.

[0240] Specifically, this application sets different numerical ranges for different cumulative collection volumes and collection growth rates, each numerical range corresponds to a different score, and the two scores respectively reflect the authority and attention of the matching journal.

[0241] Step S7142: Add the first score and the second score to obtain the collection popularity score.

[0242] In some embodiments, adjusting the ranking of each matching journal in the recommended journal list according to the popularity ranking list includes:

[0243] The average ranking of the matching journals in the popularity ranking list and the recommended journal list is taken as the new ranking of the matching journals in the recommended journal list. Specifically, for example, if a matching journal is ranked 1 in the popularity ranking list and 9 in the recommended journal list, the ranking is adjusted to 5 after taking the average; if two matching journals have the same ranking after averaging, they can be randomly sorted or user preferences can be determined based on the historical browsing data of the user account, such as impact factor preference, geographic location preference, etc., and the journal that better meets the user's preference will be placed in a higher position among the two matching journals with the same ranking.

[0244] See also Figure 8 In some embodiments, the method further comprises:

[0245] Step S81, obtaining the number of clicks of each matching journal, and modifying the corresponding matching degree based on the number of clicks.

[0246] Step S82, training a journal matching model based on the modified matching degree and the corresponding journal feature vector.

[0247] Specifically, after displaying the list of recommended journals, the number of clicks by users on each matching journal is obtained, the number of clicks and the matching degree are added together to obtain a new matching degree, and the journal matching model is feedback trained based on the new matching degree. In this way, self-optimization is performed based on user feedback, and model parameters are continuously iterated to improve matching accuracy.

[0248] See also Fig. 9 Another embodiment of the present application provides a journal matching recommendation device based on big data, comprising:

[0249] The target document module 101 is used to obtain the document to be detected and extract feature information therein to form a target feature vector.

[0250] The journal module 102 is used to obtain historical journal data and extract characteristic information of each journal to form a corresponding journal characteristic vector.

[0251] The training module 103 is used to train the machine learning model based on the feature vectors of each journal to obtain a journal matching model.

[0252] The matching module 104 is used to input the target feature vector into the journal matching model to obtain multiple matching journals and corresponding matching degrees.

[0253] The calculation module 105 is used to calculate the similarity between the target feature vector and the journal feature vectors of each matching journal.

[0254] The recommendation module 106 is used to sort the matching journals based on similarity and matching degree to obtain a recommended journal list.

[0255] The above embodiment provides a journal matching recommendation device based on big data, which trains the journal matching model through the training module 103 to obtain multiple matching journals and corresponding matching degrees, and then calculates the similarity between the target feature vector of the document to be detected and the journal feature vector of each matching journal through the calculation module 105, and finally forms a recommended journal list by combining the similarity and the matching degree. The above device combines the similarity and the matching method of the large model based on the feature vector, which not only avoids the situation where the model output is affected by the inaccurate labels set by humans, greatly improves the accuracy of journal matching, but also reduces the amount of data required for similarity calculation, that is, the document to be detected only needs to calculate the similarity with the matching journal output by the model, which improves the matching speed.

[0256] Furthermore, the device also includes a preprocessing module, which is used to remove preset stop words in the document to be detected before extracting the feature information of the document to be detected; and to perform word segmentation, stem restoration and lowercase conversion on the document to be detected.

[0257] See also Fig.10 Furthermore, the target document module 101 includes:

[0258] The emotion unit 11 is used to extract the emotion feature vector of the document to be detected based on a preset emotion dictionary.

[0259] The word frequency unit 12 is used to extract the word frequency feature vector of the document to be detected by using the TF-IDF method.

[0260] The combining unit 13 is used to combine the sentiment feature vector and the word frequency feature vector to obtain a target feature vector.

[0261] Furthermore, the emotion unit 11 is used for:

[0262] Assign a sentiment polarity score to each word in the document to be detected according to the preset sentiment dictionary.

[0263] The document sentiment score of the document to be detected is calculated based on each sentiment polarity score.

[0264] The emotion type and emotion intensity of the document to be detected are determined according to the document emotion score.

[0265] The document sentiment score, sentiment type and sentiment intensity are constructed as a sentiment feature vector.

[0266] Furthermore, the device also includes:

[0267] A statistics module is used to calculate the chi-square statistic of each journal feature vector before training the machine learning model.

[0268] The contingency table module is used to construct the contingency table corresponding to each journal feature vector.

[0269] The threshold module is used to determine the chi-square threshold of the journal feature vector based on the preset significance level and contingency table.

[0270] The elimination module is used to eliminate the corresponding journal feature vector if the chi-square statistic is less than the chi-square threshold.

[0271] Furthermore, the threshold module includes:

[0272] The degrees of freedom unit is used to reduce the number of rows and columns in the contingency table by 1 respectively, and then multiply them to obtain the degrees of freedom.

[0273] The threshold unit is used to multiply the preset significance level and the degrees of freedom to obtain the chi-square threshold.

[0274] Furthermore, the calculation module 105 is used to calculate the Mankard distance or Jaccard similarity coefficient between the target feature vector and the journal feature vector, and use the Mankard distance or Jaccard similarity coefficient as the similarity.

[0275] See also Fig.11, further, the recommendation module 106 includes:

[0276] The normalization unit 61 is used to normalize each similarity and each matching degree respectively.

[0277] The ranking value unit 62 is used to add the normalized similarity and matching degree to obtain the ranking value of the matching journal.

[0278] The recommendation list unit 63 is used to sort the matching journals according to the sorting value to obtain a recommended journal list.

[0279] Furthermore, the device also includes:

[0280] The popularity ranking module is used to calculate the popularity score of each matching journal based on the knowledge graph method and obtain a popularity ranking list.

[0281] The sorting update module is used to adjust the sorting of each matching journal in the recommended journal list according to the popularity sorting list.

[0282] Furthermore, the heat ranking module includes:

[0283] The graph construction unit is used to take each matching journal as a node, the journal feature vector as the attribute of the corresponding node, and the citation relationship between each matching journal as the edge connecting each node to obtain a journal knowledge graph.

[0284] The comment heat unit is used to obtain the comment heat score according to the comment text of each node and the corresponding comment time.

[0285] The collection heat unit is used to obtain the collection heat score according to the cumulative collection volume and collection growth rate of each node.

[0286] The node score unit is used to perform weighted calculation on the comment heat score and the collection heat score to obtain the node score.

[0287] The node heat unit is used to calculate the heat score of each node according to the score of each node and the edges between the nodes using a graph algorithm.

[0288] The heat ranking unit is used to sort the matching journals corresponding to each node according to the heat score to obtain a heat ranking list.

[0289] Furthermore, the collection heat unit is used to:

[0290] Determine a first score of the collection interval where the cumulative collection amount is located, and a second score of the growth rate interval where the collection growth rate is located.

[0291] Add the first score and the second score to get the collection popularity score.

[0292] Furthermore, the ranking update module is used to obtain the average ranking of the matching journals in the popularity ranking list and the recommended journal list as a new ranking of the matching journals in the recommended journal list.

[0293] Furthermore, the device also includes:

[0294] The click acquisition module is used to obtain the number of clicks for each matching journal and modify the corresponding matching degree based on the number of clicks.

[0295] A feedback training module is used to train the journal matching model based on the modified matching degree and the corresponding journal feature vector.

[0296] For the specific limitations of a journal matching recommendation device based on big data provided in this embodiment, please refer to the embodiment of a journal matching recommendation method based on big data above, which will not be repeated here. Each module in the above-mentioned journal matching recommendation device based on big data can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0297] In the embodiments disclosed in the present application, it should be understood that the disclosed products can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of modules, which can be electrical, mechanical or other forms. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, each functional module in each embodiment of the present invention can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0298] An embodiment of the present application provides a computer device, which may include a processor, a memory, a network interface, and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the processor executes the steps of a journal matching recommendation method based on big data as in any of the above embodiments.

[0299] The working process, working details and technical effects of the computer device provided in this embodiment can be found in the above embodiment of a journal matching recommendation method based on big data, which will not be described in detail here.

[0300] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a journal matching recommendation method based on big data as in any of the above embodiments are implemented. The computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive and / or a memory stick, etc. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0301] The working process, working details and technical effects of the computer-readable storage medium provided in this embodiment can be found in the above embodiment of a journal matching recommendation method based on big data, which will not be described in detail here.

[0302] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0303] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0304] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A journal matching recommendation method based on big data, characterized in that: include: Obtain the document to be detected, extract the feature information therein to form a target feature vector; Obtain historical journal data, extract characteristic information of each journal to form a corresponding journal characteristic vector; Training a machine learning model based on the feature vectors of each journal to obtain a journal matching model; Inputting the target feature vector into the journal matching model to obtain a plurality of matching journals and corresponding matching degrees; Calculating the similarity between the target feature vector and the journal feature vector of each matching journal; The matching journals are sorted based on the similarity and the matching degree to obtain a recommended journal list.

2. The journal matching recommendation method based on big data according to claim 1 is characterized in that: The step of obtaining a document to be detected and extracting feature information therein to form a target feature vector includes: Extracting the emotional feature vector of the document to be detected based on a preset emotional dictionary; The TF-IDF method is used to extract the word frequency feature vector of the document to be detected; The sentiment feature vector and the word frequency feature vector are combined to obtain the target feature vector.

3. The journal matching recommendation method based on big data according to claim 2 is characterized in that: The step of extracting the emotional features of the document to be detected based on a preset emotional dictionary includes: Assigning a sentiment polarity score to each word in the document to be detected according to the preset sentiment dictionary; Calculating the sentiment score of the document to be detected based on each of the sentiment polarity scores; Determine the emotion type and emotion intensity of the document to be detected according to the document emotion score; The document sentiment score, the sentiment type and the sentiment intensity are constructed into the sentiment feature vector.

4. The journal matching recommendation method based on big data according to claim 1 is characterized in that: Also includes: Before training the machine learning model, calculating the chi-square statistic of each of the journal feature vectors; Constructing a contingency table corresponding to each of the journal feature vectors; wherein the rows of the contingency table represent the values ​​of different features in the journal feature vector, and the columns represent the categories of different features; Determining a chi-square threshold of the journal feature vector according to a preset significance level and the contingency table; If the chi-square statistic is less than the chi-square threshold, the corresponding journal feature vector is eliminated.

5. The journal matching recommendation method based on big data according to claim 1 is characterized in that: Also includes: Calculate the popularity score of each matching journal based on the knowledge graph method to obtain a popularity ranking list; The ranking of each matching journal in the recommended journal list is adjusted according to the popularity ranking list.

6. The journal matching recommendation method based on big data according to claim 5 is characterized in that: The method based on the knowledge graph calculates the popularity score of each matching journal to obtain a popularity ranking list, including: Taking each of the matching journals as a node, taking the journal feature vector as an attribute corresponding to the node, and taking the citation relationship between the matching journals as an edge connecting the nodes, to obtain a journal knowledge graph; Obtaining a comment popularity score according to the comment text and corresponding comment time of each node; The collection popularity score is obtained according to the accumulated collection amount and collection growth rate of each node; Performing weighted calculation on the comment heat score and the collection heat score to obtain a node score; Calculate the heat score of each node according to the node score and the edges between the nodes using a graph algorithm; The matching journals corresponding to each of the nodes are sorted according to the popularity scores to obtain the popularity sorted list.

7. The journal matching recommendation method based on big data according to claim 6 is characterized in that: The collection heat score is obtained according to the accumulated collection amount and collection growth rate of each node, including: Determine a first score of the collection interval in which the cumulative collection amount is located, and a second score of the growth rate interval in which the collection growth rate is located; The first score and the second score are added together to obtain the collection popularity score.

8. The journal matching recommendation method based on big data according to claim 5 is characterized in that: The step of adjusting the ranking of each matching journal in the recommended journal list according to the popularity ranking list includes: The average of the rankings of the matching journals in the popularity ranking list and the recommended journal list is taken as the new ranking of the matching journals in the recommended journal list.

9. A journal matching recommendation device based on big data, characterized in that: include: The target document module is used to obtain the document to be detected and extract the feature information therein to form a target feature vector; The journal module is used to obtain historical journal data and extract the characteristic information of each journal to form the corresponding journal feature vector; A training module, used for training a machine learning model based on each of the journal feature vectors to obtain a journal matching model; A matching module, used for inputting the target feature vector into the journal matching model to obtain a plurality of matching journals and corresponding matching degrees; A calculation module, used for calculating the similarity between the target feature vector and the journal feature vector of each matching journal; The recommendation module is used to sort the matching journals based on the similarity and the matching degree to obtain a recommended journal list.

10. A computer device 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 computer program, the steps of the journal matching recommendation method based on big data as described in any one of claims 1 to 8 are implemented.

Images