Paper data availability classification method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a paper data availability classification method and device, equipment and a storage medium. When the method is executed, a to-be-classified paper text is firstly obtained, early-stage data processing is carried out on the to-be-classified paper text, and a paper data availability declaration text is extracted; secondly, inputting the paper data availability declaration text into a data availability classification model; the method comprises the following steps: preprocessing a paper data availability declaration text through a data availability classification model to obtain a vector sequence, carrying out convolution operation on the vector sequence to obtain text features, mapping the text features into a preset classification set, and outputting a classification result of data availability in the to-be-classified paper text. The data availability in the papers is classified, and a basis is provided for follow-up database construction according to data availability classification.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a method, device, equipment and storage medium for classifying paper data availability. Background Art

[0002] In academic papers, the "availability" of data is an important part of the international FAIR data principles and is crucial for data sharing and data application. Especially in the massive biomedical papers, a large number of studies rely on the use and analysis of data, which are usually referred to as scientific paper associated data. However, not all papers cite data in a standardized manner or explain the use of data, especially in terms of annotating and explaining the accessibility and availability of data.

[0003] In order to improve the accessibility and usability of data associated with papers, the five major international publishing groups, including Elsevier, Springer, Willey, Taylor & Francis and Sage, have formulated data policies at different levels and require the writing of Data Availability Statements (DAS) in papers. These publishing groups require their journals to regulate the data availability in papers according to different data policies.

[0004] However, due to the differences in data availability statements among different publishing groups and the lack of unified grading standards, description methods and formats, it is difficult to integrate data citation information in scientific literature and it is also impossible to accurately classify the data availability in papers. Summary of the invention

[0005] In view of this, the embodiments of the present application provide a method, device, equipment and storage medium for classifying paper data availability, aiming to solve the problem that data availability in papers cannot be accurately classified.

[0006] In a first aspect, an embodiment of the present application provides a method for classifying paper data availability, the method comprising:

[0007] Get the text of the paper to be classified;

[0008] Performing data processing on the paper text to be classified, and extracting the paper data availability statement text, wherein the paper data availability statement text is the text in the paper used to describe the availability of paper data;

[0009] The data availability declaration text of the paper is used as the input of the data availability classification model to obtain an availability classification result; wherein the data availability classification model is used to preprocess the data availability declaration text of the paper and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the data availability declaration text of the paper, and map the text features to a preset classification collection to obtain the availability classification result of the data availability declaration text of the paper.

[0010] In a possible implementation, the data availability classification model includes a first module and a second module;

[0011] The first module is used to preprocess the data availability statement text of the paper and convert it into a vector sequence;

[0012] The second module is used to perform a convolution operation on the vector sequence to obtain text features of the paper data availability statement text, and map the text features to a preset classification collection to obtain the availability classification result of the paper data availability statement text.

[0013] In a possible implementation, preprocessing the paper data availability statement text and converting it into a vector sequence includes:

[0014] Segmenting the data availability statement text of the paper to obtain multiple sentence text sequences;

[0015] Perform word segmentation, stop word removal and word embedding processing on each of the sentence text sequences in sequence to obtain a plurality of word embedding vectors, wherein the word embedding vectors correspond one-to-one to the sentence text sequences;

[0016] The multiple word embedding vectors are processed based on a self-attention mechanism and position encoding to obtain the vector sequence.

[0017] In a possible implementation, performing a convolution operation on the vector sequence to obtain text features of the paper data availability statement text includes:

[0018] Performing a convolution operation on the vector sequence through a convolution kernel to extract multiple local features of the paper data availability statement text;

[0019] The multiple local features are spatially downsampled, and the multiple local features are aggregated into the text feature.

[0020] In a possible implementation, mapping the text features to a preset classification collection to obtain the availability classification result of the paper data availability statement text includes:

[0021] The text features are mapped to the preset classification collection through the activation function of the fully connected layer to obtain the availability classification result, and the number of neurons in the fully connected layer is the same as the number of categories in the preset classification collection.

[0022] In one possible implementation, the preset classification collection includes data that is completely not public, data that needs to be obtained by contacting the author or applying to the data repository, data that is provided in the text or in an attachment, data that is partially included in the text, and data that is completely public in the public data repository.

[0023] In a possible implementation, the training method of the data availability classification model is:

[0024] Obtaining a sample text with a data availability classification label, where the sample text is a data availability statement text of a paper;

[0025] Using the model to be trained to perform data availability classification on the sample text to obtain a data availability classification result;

[0026] According to the data availability classification result and the data availability classification label, the parameters of the model to be trained are optimized to obtain the data availability classification model; the model to be trained is used to preprocess the sample text and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the sample text, and the text features are mapped to a preset classification collection to obtain the data availability classification result of the sample text.

[0027] In a second aspect, an embodiment of the present application provides a device for classifying paper data availability, the device comprising:

[0028] The acquisition module is used to obtain the text of the paper to be classified;

[0029] A processing module, used for performing data processing on the paper text to be classified, and extracting the paper data availability declaration text, wherein the paper data availability declaration text is the text in the paper used to describe the availability of paper data;

[0030] A classification module is used to use the data availability declaration text of the paper as the input of a data availability classification model to obtain an availability classification result; wherein the data availability classification model is used to preprocess the data availability declaration text of the paper and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the data availability declaration text of the paper, and map the text features to a preset classification collection to obtain the availability classification result of the data availability declaration text of the paper.

[0031] In a possible implementation, the data availability classification model includes a first module and a second module;

[0032] The first module is used to preprocess the data availability statement text of the paper and convert it into a vector sequence;

[0033] The second module is used to perform a convolution operation on the vector sequence to obtain the text features of the paper data availability statement text, and map the text features to a preset classification collection to obtain the availability classification result of the paper data availability statement text.

[0034] In a possible implementation manner, the first module is specifically configured to:

[0035] Segmenting the data availability statement text of the paper to obtain multiple sentence text sequences;

[0036] Perform word segmentation, stop word removal and word embedding processing on each of the sentence text sequences in sequence to obtain a plurality of word embedding vectors, wherein the word embedding vectors correspond one-to-one to the sentence text sequences;

[0037] The multiple word embedding vectors are processed based on a self-attention mechanism and position encoding to obtain the vector sequence.

[0038] In a possible implementation, the second module is specifically used to: perform a convolution operation on the vector sequence through a convolution kernel to extract multiple local features of the paper data availability statement text;

[0039] The multiple local features are spatially downsampled, and the multiple local features are aggregated into the text feature.

[0040] In a possible implementation, the second module is specifically used to map the text features to the preset classification collection through an activation function of a fully connected layer to obtain the availability classification result, and the number of neurons in the fully connected layer is the same as the number of categories in the preset classification collection.

[0041] In one possible implementation, the preset classification collection includes data that is completely not public, data that needs to be obtained by contacting the author or applying to the data repository, data that is provided in the text or in an attachment, data that is partially included in the text, and data that is completely public in the public data repository.

[0042] In a possible implementation, the training method of the data availability classification model is:

[0043] Obtaining a sample text with a data availability classification label, where the sample text is a data availability statement text of a paper;

[0044] Using the model to be trained to perform data availability classification on the sample text to obtain a data availability classification result;

[0045] According to the data availability classification result and the data availability classification label, the parameters of the model to be trained are optimized to obtain the data availability classification model; the model to be trained is used to preprocess the sample text and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the sample text, and the text features are mapped to a preset classification collection to obtain the data availability classification result of the sample text.

[0046] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the computer device executes the paper data availability classification method described in any one of the first aspects above.

[0047] In a fourth aspect, an embodiment of the present application provides a computer storage medium, in which a code is stored. When the code is executed, the device executing the code implements the paper data availability classification described in any one of the first aspects above.

[0048] The embodiment of the present application provides a method, device, equipment and storage medium for classifying data availability of papers. When executing the method, the paper text to be classified is first obtained, and the preliminary data processing is performed on the paper text to be classified to extract the paper data availability declaration text; then, the paper data availability declaration text is input into the data availability classification model; the paper data availability declaration text is preprocessed by the data availability classification model to obtain a vector sequence, and the vector sequence is convoluted to obtain text features, and then mapped to a preset classification collection according to the text features, and the classification result of the data availability in the paper text to be classified is output, so as to realize the classification of data availability in the paper, and provide a basis for the subsequent construction of a database according to the data availability classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A flowchart of a method for classifying paper data availability provided in an embodiment of the present application;

[0051] Figure 2 A structural diagram of a data availability classification model provided in an embodiment of the present application;

[0052] Figure 3 A schematic diagram of a data availability classification model training process provided in an embodiment of the present application;

[0053] Figure 4 A schematic diagram of the structure of a device for classifying paper data availability provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] At present, the research directions on data reusability at home and abroad are mainly focused on the factors affecting data reuse, related ethics and policies, etc., and there are few studies on the identification of the availability of data in scientific literature.

[0055] In view of this problem, the embodiment of the present application provides a method, device, equipment and storage medium for classifying data availability of papers. When executing the method, the paper text to be classified is first obtained, and the paper text to be classified is subjected to preliminary data processing to extract the paper data availability declaration text; then, the paper data availability declaration text is input into the data availability classification model; the paper data availability declaration text is preprocessed by the data availability classification model to obtain a vector sequence, and the vector sequence is convoluted to obtain text features, and then mapped to a preset classification collection according to the text features, and the classification result of the data availability in the paper text to be classified is output, so as to realize the classification of data availability in the paper, and provide a basis for the subsequent construction of a database according to the data availability classification.

[0056] Obviously, the embodiments described in this application are only part of the embodiments of this application, rather than all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0057] See also Figure 1 , Figure 1 A flowchart of a method for classifying paper data availability provided in an embodiment of the present application includes:

[0058] S101: Obtain the paper text to be classified.

[0059] The paper text to be classified refers to the paper that needs to classify the data in the paper. The paper text to be classified includes the text used to describe the data availability, that is, the paper data availability statement.

[0060] These papers may not adequately annotate or describe the accessibility and usability of data, and need to be classified and labeled according to relevant data policies and guidelines. By classifying data availability, data in papers can be better managed and organized to improve data availability and reproducibility, and promote transparency and sustainability of scientific research.

[0061] S102: Perform data processing on the paper text to be classified and extract the paper data availability statement text.

[0062] The text of the paper data availability statement is the text used in the paper to describe the availability of the paper data. The Data Availability Statement (DAS) is a declaration or description method of data availability in the paper. Specifically, it refers to a paragraph provided by the author in the paper to explain the accessibility and reusability of the data used in the research. The text of the paper data availability statement helps readers understand and evaluate the availability of the data used in the paper, and provides more transparency and traceability for scientific research. At the same time, for academic publishing organizations and research institutions, encouraging and requiring authors to provide data availability statements will also help standardize data management and data sharing practices.

[0063] The data availability statement of the paper may include the following contents:

[0064] Data acquisition channel: Describe the source of the data and the method of obtaining it, such as public databases, provided by partners, etc.

[0065] Data usage permission: describes the data usage permission, including data copyright information, usage restrictions, etc.

[0066] Data format: Provides a description of the data format, such as data file type, storage format, etc.

[0067] Data reusability: Describe the data’s potential for reuse and possible limitations, including the data’s availability period, related conditions, etc.

[0068] By classifying and labeling the data availability statements in papers, researchers and publishers can better understand the availability of data in papers, thereby promoting the reasonable use and sharing of data. By identifying and labeling the data availability statements in these papers, more normative guidance and support can be provided for data management and scientific research practices.

[0069] Since the position of the paper data availability statement text in the paper is relatively fixed, it usually appears at the end of the paper, that is, after the reference list and before the end of the text. In the embodiment of the present application, the paper data availability statement text can be extracted from the paper text to be classified by locating the position of the paper data availability statement text in the paper through a text extraction model.

[0070] S103: Using the paper data availability declaration text as input to a data availability classification model to obtain an availability classification result.

[0071] In an embodiment of the present application, a data availability classification model is used to preprocess the data availability statement text of the paper and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the data availability statement text of the paper, and map the text features to a preset classification collection to obtain an availability classification result of the data availability statement text of the paper.

[0072] First, preprocess the data availability statement text of the paper, such as cleaning, tokenizing, and removing stop words, to prepare the data for subsequent processing. Secondly, perform text vectorization, convert the preprocessed text into a vector sequence, and represent the text data with numbers. You can use the Bag of Words model or the Word Embedding model such as Word2Vec, GloVe, etc. Then, perform a convolution operation on the vector sequence, and use a convolutional neural network (CNN) to extract local features in the text. The convolution operation helps capture local correlations in the text. Finally, perform feature mapping and classification, and map the text features obtained by the convolution operation to a preset classification collection. For example, the text features are reduced in dimension through a fully connected layer, and the softmax function is used for multi-classification to obtain the availability classification results of the data availability statement text of the paper.

[0073] It should be noted that the data availability classification model can be adjusted and optimized according to specific tasks and data. For example, different text processing techniques can be used, and other deep learning models can be introduced to improve classification performance.

[0074] In one possible implementation, the data availability classification model includes a first module and a second module;

[0075] The first module is used to preprocess the data availability statement text of the paper and convert it into a vector sequence; the second module is used to perform a convolution operation on the vector sequence to obtain the text features of the data availability statement text of the paper, map the text features to the preset classification collection, and obtain the availability classification results of the data availability statement text of the paper.

[0076] In the embodiment of the present application, the first module is used to implement text preprocessing and vector sequence conversion, which usually includes preprocessing steps such as text cleaning, word segmentation, and removal of stop words, and then converts the text into a vector sequence. Common methods include bag of words model (Bag of Words), TF-IDF (Term Frequency-Inverse Document Frequency), word embedding model (such as Word2Vec, GloVe), etc. The first module can convert text data into a vector form that can be processed by the machine learning model, preparing for subsequent processing.

[0077] The second module is used to implement the convolution operation and classification module. It is mainly used to perform convolution operations on vector sequences, extract local features of text, and then classify according to text features. In deep learning models, convolutional neural networks (CNNs) are often used for text classification tasks. They can effectively capture local information in text and obtain text features. Then, through the fully connected layer and softmax function, classification is performed, and the text features are mapped to the preset classification collection to obtain the final text classification result.

[0078] In one possible implementation, the first module is a biological entity recognition and relationship extraction pre-training (BioBert) model, and the second module is a text convolutional neural network (TextCNN) model.

[0079] The BioBert model is a pre-trained language model specifically designed for processing text in the biomedical field. It is obtained by pre-training on large-scale text data in the biomedical field based on BERT. The BioBert model can better capture the semantics and features of biomedical text.

[0080] The TextCNN model is a text classification model based on convolutional neural networks. It uses convolution operations to capture local features in text and integrates these features through pooling operations. Compared with traditional RNN models, TextCNN has the advantages of high computational efficiency and strong parallel computing capabilities, and is widely used in text classification tasks.

[0081] The data availability classification model in the embodiment of the present application includes a pre-trained BioBert model for text preprocessing and vector representation, and a TextCNN model for convolution operations and classification of vector sequences. This combination can effectively handle data availability classification tasks in the biomedical field, especially in utilizing domain-specific text representations and extracting local features from text.

[0082] Combine the following Figure 2The data availability classification model provided in the embodiment of the present application is introduced. Figure 2 shown.

[0083] In a possible implementation, preprocessing the paper data availability statement text and converting it into a vector sequence includes:

[0084] Segmenting the data availability statement text of the paper to obtain multiple sentence text sequences;

[0085] Perform word segmentation, stop word removal and word embedding processing on each of the sentence text sequences in sequence to obtain a plurality of word embedding vectors, wherein the word embedding vectors correspond one-to-one to the sentence text sequences;

[0086] The multiple word embedding vectors are processed based on a self-attention mechanism and position encoding to obtain the vector sequence.

[0087] The first module segments the input paper data availability statement text into multiple sentence text sequences (W 1 , W 2 , W 3 , W 4 ...W n ), and then each sentence text sequence is converted into the corresponding word embedding vector (E 1 、E 2 、E 3 、E 4 ...E n ), the word embedding vector is the representation of each word in the text by the data availability classification model, which converts each word in the text into a vector of fixed dimension.

[0088] The multiple word embedding vectors are directly input into the encoder of the first model for processing, and the encoder can be a Transformer encoder. The Transformer encoder is a deep learning network based on the Transformer structure. The Transformer encoder is designed to capture long-distance dependencies in the input sequence.

[0089] Specifically, the Transformer encoder consists of multiple Transformer layers, each of which contains multiple self-attention sub-layers and feed-forward neural network sub-layers. The self-attention mechanism is able to adjust the weights of each word embedding vector according to the importance of each word in the input sequence, thereby focusing more attention on key information. Position encoding is used to add position information to each input vector to help the model understand the relative order in the input sequence.

[0090] Multiple word embedding vectors are directly input into the first layer of the Transformer encoder for processing. After being processed by the Transformer encoder, the output is a high-dimensional vector sequence (T 1 , T 2 , T 3 , T 4 ...T n , T [CLS] and T [SEP] is a special token identifier used to indicate the starting and ending positions in the sequence). The vector sequence will contain the key semantic and contextual information in the input sequence, which can be used for subsequent processing and tasks, such as feature extraction and classification.

[0091] In the embodiment of the present application, NLP tools such as NLTK (Natural Language Toolkit) can be used to segment long text data. Segmentation is the process of splitting long text into words or phrases, making text data easier to process.

[0092] In the embodiment of the present application, stop words are common words that have no practical analytical value in text processing, such as "the", "for", "and", etc. A stop word list is used to reduce data noise and improve the efficiency of text vectorization by removing stop words.

[0093] In a possible implementation, performing a convolution operation on the vector sequence to obtain text features of the paper data availability statement text includes:

[0094] Performing a convolution operation on the vector sequence through a convolution kernel to extract multiple local features of the paper data availability statement text;

[0095] The multiple local features are spatially downsampled, and the multiple local features are aggregated into the text feature.

[0096] The second module performs convolution operations on the vector sequences output by the first module to obtain local features of the data availability statement text, and then obtains text features through spatial downsampling and aggregation. The second module can use the TextCNN model to further process these output vector sequences. The TextCNN model is a natural language processing model based on convolutional neural networks (CNNs) and is suitable for tasks such as text classification and named entity recognition.

[0097] In this step, the TextCNN model uses a convolutional layer to capture local features in the vector sequence, which can play a key role in text classification tasks. This layer uses a series of convolution kernels (or filters) to perform convolution operations on the input vector sequence. The convolution operation can capture local text features. The input vector sequence is expanded into a one-dimensional matrix and convolution operations are performed on it using convolution kernels to extract local features in the text. The TextCNN model can extract local features at different positions and scales, which can help the model better understand the semantics and content of the text.

[0098] After the convolution operation, the spatial downsampling (pooling operation) is performed through the pooling layer to reduce the dimension of the feature map, thereby reducing the number of model parameters while retaining the most important feature information. The most common downsampling operations are maximum pooling or average pooling, which can perform corresponding operations on each feature map to obtain a smaller feature representation.

[0099] Finally, the aggregation operation combines multiple local features obtained through convolution and downsampling into text features, which can be used for subsequent text classification or other tasks. Text features can include length distribution, syntactic features, grammatical features, and word usage features.

[0100] In the embodiment of the present application, the convolution operation is fully utilized to extract the local features of the text, and more global text features are obtained through downsampling and aggregation, thereby providing richer feature representation for the data availability classification task.

[0101] In a possible implementation, mapping the text features to a preset classification collection to obtain the availability classification result of the paper data availability statement text includes:

[0102] The text features are mapped to the preset classification collection through the activation function of the fully connected layer to obtain the availability classification result, and the number of neurons in the fully connected layer is the same as the number of categories in the preset classification collection.

[0103] In the fully connected layer, an activation function (usually a softmax function) can be used to map text features to a preset classification set. The softmax function can map the input vector to a probability distribution, where each category corresponds to a probability value. Therefore, for data availability classification tasks, the number of neurons in the fully connected layer is usually the same as the number of categories in the preset classification set, so as to generate corresponding probabilities for each category.

[0104] After processing by the fully connected layer, a probability distribution can be obtained, which represents the possibility of different classification results of paper data availability. According to this probability distribution, the category with the highest probability can be selected as the final classification result, or multiple classification results can be determined according to a preset threshold.

[0105] In one possible implementation, the preset classification collection includes data that is completely not public, data that needs to be obtained by contacting the author or applying to the data repository, data that is provided in the text or in an attachment, data that is partially included in the text, and data that is completely public in the public data repository.

[0106] This application embodiment refers to the data sharing classification policies of major journal publishers and institutions, the requirements of data citation elements, and the existing data reuse practice characteristics summarized in previous studies, and combines the analysis of the characteristics of information related to the reusability of biomedical paper-related data to perform fine-grained data reusability classification and formulate classification standards. Combined with the data sharing policies of the five major publishers, according to the degree of data accessibility and the method of access, the preset classification collection includes:

[0107] (1) The data is not disclosed at all, for example: NONEreported;

[0108] (2) Access can be obtained by contacting the author or applying to the data repository, for example: Access requests can be submitted to Prof. JM Wardlaw;

[0109] (3) The data are available in the article or in the supplementary data files, for example: The data generated in this study are available within the article and its Supplementary data files;

[0110] (4) Data are partially included in the paper, partially public / available on request, or only the data storage institution is published, for example: All data and code compiled for this study are available on Github. (no URL);

[0111] (5) The data are fully disclosed in public data repositories, for example: The Data for this study have been deposited in GEO with accession code: GSE 114234.

[0112] Through the paper data availability classification method provided by the embodiment of the present application. First, the paper text to be classified is obtained, and the preliminary data processing is performed on the paper text to be classified to extract the paper data availability declaration text; then, the paper data availability declaration text is input into the data availability classification model; the paper data availability declaration text is preprocessed by the data availability classification model to obtain a vector sequence, and the vector sequence is convoluted to obtain text features, and then mapped to a preset classification collection according to the text features, and the classification result of the data availability in the paper text to be classified is output, so as to realize the classification of data availability in the paper, and provide a basis for the subsequent construction of a database according to the data availability classification.

[0113] The present application embodiment also provides a method for training a data availability classification model, specifically:

[0114] Obtaining a sample text with a data availability classification label, where the sample text is a data availability statement text of a paper;

[0115] Using the model to be trained to perform data availability classification on the sample text to obtain a data availability classification result;

[0116] According to the data availability classification result and the data availability classification label, the parameters of the model to be trained are optimized to obtain the data availability classification model; the model to be trained is used to preprocess the sample text and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the sample text, and the text features are mapped to a preset classification collection to obtain the data availability classification result of the sample text.

[0117] The paper data in the embodiments of this application can be selected from the full-text database of biomedical and life science journals PubMed (PMC). The present application example uses open text data mining in PMC, biomedical papers in XML format annotated with HTML language as the data source, processes English text, and uses Python to extract the paper data availability statement text from the full text of the document as the sample text for model training.

[0118] In one possible implementation, before using the sample text to train the model, the sample text can also be deduplicated to ensure the uniqueness of the data set and remove data with a text length greater than 2,000 characters and data that is irrelevant to data availability, which helps reduce noise and improve data quality.

[0119] In a possible implementation, the accuracy P can be used i , recall rate R i, F1 value and other machine learning model evaluation indicators to evaluate the classification model, and use expert consultation to comprehensively evaluate the classification standards and classification identification methods. The specific formula is as follows:

[0120]

[0121] Among them, C i is the classification category, C i The data may be completely private, available by contacting the author or applying to a data repository, provided in the article or in an attachment, partially included in the article, or fully available in a public data repository.

[0122] In an embodiment of the present application, the model to be trained can be a machine learning model, including a naive Bayes (NB) model, a support vector machine (SVM) model, a logistic regression (LR) model, a random forest (RF) model, an adaptive reinforcement learning (AdaBoost) model, and a K nearest neighbor (KNN) model; a deep learning model can also be selected, including a text convolutional neural network (TextCNN) model, a fast text (FastText) model, a biological entity recognition and relationship extraction pre-training (BioBert) model, and a model-combined BioBert_TextCNN model.

[0123] like Figure 3 As shown, biomedical papers in XML format are selected from PMC, and the biomedical papers are preprocessed and labeled to obtain sample texts with data availability classification labels as a data set. The data set is then input into any model to be trained to finally obtain the corresponding availability classification results.

[0124] Next, each model to be trained is introduced.

[0125] The Naive Bayes (NB) model is based on the Bayesian theorem. Naive Bayes assumes that features are independent of each other. It calculates the probability of each category and then classifies based on the conditional probability of the feature. The structure of the model includes calculating the prior probability of each category and the conditional probability of each feature under a given category. Through these probabilities, the probability of the text belonging to each category can be calculated, and the category with the highest probability is selected as the final classification result.

[0126] Support Vector Machine (SVM) model, the core idea of ​​SVM is to find an optimal hyperplane to maximize the interval between data points of different categories, so as to achieve linear classification of data. SVM finds a hyperplane through mathematical optimization, where the support vector is the data point closest to the hyperplane. Support vector machines can use different kernel functions to handle nonlinear classification problems.

[0127] Logistic regression (LR) model, logistic regression handles binary classification problems, it simulates the probability of output data being a certain category. It uses the sigmoid function to convert the linear combination features into probability values. The structure of the model includes the input layer, linear combination, and sigmoid function, and the output probability value represents the probability that the data point belongs to the positive category.

[0128] Random Forest (RF) Model,Random Forest is an integrated learning model consisting of multiple decision trees. Each decision tree is trained independently on the data, and then the final result is obtained by voting or averaging. Random Forest includes multiple decision trees, each of which is trained by randomly extracting samples and features. Finally, classification is performed by combining multiple decision trees.

[0129] Adaptive reinforcement learning (AdaBoost) model, AdaBoost is an ensemble learning method that builds a strong classifier by combining multiple weak classifiers. The weight of each weak classifier is adjusted according to the error of the previous classifier. The structure of AdaBoost includes multiple weak classifiers, each of which adjusts the weight of the sample to correct the importance of the misclassified sample, and finally a strong classifier is obtained by combining.

[0130] K-Nearest Neighbor (KNN) Model, K-Nearest Neighbor is a non-parametric method that classifies based on the votes of neighbors. It assumes that nearby data points have similar labels. The structure of the model is simple, it calculates the closest K data points and then classifies by majority voting.

[0131] Text Convolutional Neural Network (TextCNN) model, TextCNN uses convolutional neural networks to process text data and capture local features in text. Through convolution and pooling operations, the semantic information of the text is extracted. The structure of TextCNN includes input layer, convolution layer, pooling layer, and fully connected layer. The convolution layer is used to extract local features, the pooling layer is used to summarize features, and finally the fully connected layer is used for classification.

[0132] FastText model, FastText is a text classification model that uses a single-layer neural network structure of CBOW. It can quickly classify text, especially for large-scale text data. The structure of FastText includes an input layer, a hidden layer, and an output layer. The input is the embedding of all words in the text, and the labels of the output layer are obtained after passing through the hidden layer.

[0133] BioBERT is a variant of BERT specifically designed for processing biomedical text. It can understand domain-specific terminology and language and is suitable for text analysis in the biomedical field. BioBERT uses the model structure of BERT, but uses biomedical text data to train the model during the pre-training process.

[0134] BioBert_TextCNN model, BioBERT_TextCNN is a combination of the BioBERT model and the TextCNN model, and is used for text classification in the biomedical field. It fully utilizes the advantages of deep learning and convolutional neural networks. The structure of BioBERT_TextCNN includes the pre-trained model of BioBERT and the text classification structure of TextCNN, and is used for text classification tasks in the biomedical field.

[0135] In the embodiment of the present application, the same sample text is used to train the above models to obtain the corresponding models. Table 1 shows the performance comparison of the classification algorithms of each model, where Accuracy refers to the proportion of samples correctly predicted by the model on the test data set; Precision refers to the proportion of samples predicted as positive by the classification model that are actually positive; Recall refers to the proportion of samples correctly predicted as positive by the classification model among all true positive samples; F1-score is an indicator that comprehensively measures the performance of the classification model, which combines precision and recall. The F1-score provides an indicator for comprehensively evaluating the performance of the model by balancing precision and recall.

[0136] Table 1: Classification algorithm performance comparison table

[0137]

[0138] From the data shown in Table 1, we can see that the BioBERT_TextCNN model has the best overall performance, reaching an accuracy of 0.881 and a recall of 0.884. Therefore, using the BioBERT_TextCNN model as the training model to obtain a data availability classification model can improve the accuracy of the data availability classification results.

[0139] The above are some specific implementations of the paper data availability classification method provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization.

[0140] See also Figure 4The structure diagram of the paper data availability classification device 400 is shown, and the device 400 includes an acquisition module 410, a processing module 420 and a classification module 430.

[0141] An acquisition module 410 is used to acquire the text of the paper to be classified;

[0142] A processing module 420 is used to perform data processing on the paper text to be classified and extract the paper data availability declaration text, where the paper data availability declaration text is the text in the paper used to describe the availability of paper data;

[0143] The classification module 430 is used to use the paper data availability declaration text as the input of the data availability classification model to obtain an availability classification result; wherein the data availability classification model is used to preprocess the paper data availability declaration text and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the paper data availability declaration text, and map the text features to a preset classification collection to obtain the availability classification result of the paper data availability declaration text.

[0144] In a possible implementation, the data availability classification model includes a first module and a second module;

[0145] The first module is used to preprocess the data availability statement text of the paper and convert it into a vector sequence;

[0146] The second module is used to perform a convolution operation on the vector sequence to obtain the text features of the paper data availability statement text, and map the text features to a preset classification collection to obtain the availability classification result of the paper data availability statement text.

[0147] In a possible implementation manner, the first module is specifically configured to:

[0148] Segmenting the data availability statement text of the paper to obtain multiple sentence text sequences;

[0149] Perform word segmentation, stop word removal and word embedding processing on each of the sentence text sequences in sequence to obtain a plurality of word embedding vectors, wherein the word embedding vectors correspond one-to-one to the sentence text sequences;

[0150] The multiple word embedding vectors are processed based on a self-attention mechanism and position encoding to obtain the vector sequence.

[0151] In a possible implementation, the second module is specifically used to: perform a convolution operation on the vector sequence through a convolution kernel to extract multiple local features of the paper data availability statement text;

[0152] The multiple local features are spatially downsampled, and the multiple local features are aggregated into the text feature.

[0153] In a possible implementation, the second module is specifically used to map the text features to the preset classification collection through an activation function of a fully connected layer to obtain the availability classification result, and the number of neurons in the fully connected layer is the same as the number of categories in the preset classification collection.

[0154] In one possible implementation, the preset classification collection includes data that is completely not public, data that needs to be obtained by contacting the author or applying to the data repository, data that is provided in the text or in an attachment, data that is partially included in the text, and data that is completely public in the public data repository.

[0155] In a possible implementation, the training method of the data availability classification model is:

[0156] Obtaining a sample text with a data availability classification label, where the sample text is a data availability statement text of a paper;

[0157] Using the model to be trained to perform data availability classification on the sample text to obtain a data availability classification result;

[0158] According to the data availability classification result and the data availability classification label, the parameters of the model to be trained are optimized to obtain the data availability classification model; the model to be trained is used to preprocess the sample text and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the sample text, and the text features are mapped to a preset classification collection to obtain the data availability classification result of the sample text.

[0159] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0160] Among them, the device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the paper data availability classification method described in any embodiment of the present application.

[0161] The computer storage medium stores codes, and when the codes are executed, the device executing the codes implements the paper data availability classification method described in any embodiment of the present application.

[0162] The "first" and "second" in the names such as "first" and "second" (if any) mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0163] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0164] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0165] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A method for classifying paper data availability, characterized in that: include: Get the text of the paper to be classified; Performing data processing on the paper text to be classified, and extracting the paper data availability statement text, wherein the paper data availability statement text is the text in the paper used to describe the availability of paper data; The data availability declaration text of the paper is used as the input of the data availability classification model to obtain an availability classification result; wherein the data availability classification model is used to preprocess the data availability declaration text of the paper and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the data availability declaration text of the paper, and map the text features to a preset classification collection to obtain the availability classification result of the data availability declaration text of the paper.

2. The method according to claim 1, characterized in that: The data availability classification model includes a first module and a second module; The first module is used to preprocess the data availability statement text of the paper and convert it into a vector sequence; The second module is used to perform a convolution operation on the vector sequence to obtain text features of the paper data availability statement text, and map the text features to a preset classification collection to obtain the availability classification result of the paper data availability statement text.

3. The method according to claim 2, characterized in that The preprocessing of the data availability statement text of the paper and converting it into a vector sequence includes: Segmenting the data availability statement text of the paper to obtain multiple sentence text sequences; Perform word segmentation, stop word removal and word embedding processing on each of the sentence text sequences in sequence to obtain a plurality of word embedding vectors, wherein the word embedding vectors correspond one-to-one to the sentence text sequences; The multiple word embedding vectors are processed based on a self-attention mechanism and position encoding to obtain the vector sequence.

4. The method according to claim 2, characterized in that: The convolution operation is performed on the vector sequence to obtain text features of the paper data availability statement text, including: Performing a convolution operation on the vector sequence through a convolution kernel to extract multiple local features of the paper data availability statement text; The multiple local features are spatially downsampled, and the multiple local features are aggregated into the text feature.

5. The method according to claim 4, characterized in that The mapping of the text features into a preset classification collection to obtain the availability classification result of the paper data availability statement text includes: The text features are mapped to the preset classification collection through the activation function of the fully connected layer to obtain the availability classification result, and the number of neurons in the fully connected layer is the same as the number of categories in the preset classification collection.

6. The method according to any one of claims 1 to 5, characterized in that: The preset classification collection includes data that is completely not public, data that needs to be obtained by contacting the author or applying to the data repository, data that is provided in the article or in an attachment, data that is partially included in the article, and data that is completely public in the public data repository.

7. The method according to any one of claims 1 to 5, characterized in that: The training method of the data availability classification model is: Obtaining a sample text with a data availability classification label, where the sample text is a data availability statement text of a paper; Using the model to be trained to perform data availability classification on the sample text to obtain a data availability classification result; According to the data availability classification result and the data availability classification label, the parameters of the model to be trained are optimized to obtain the data availability classification model; the model to be trained is used to preprocess the sample text and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the sample text, and the text features are mapped to a preset classification collection to obtain the data availability classification result of the sample text.

8. A device for classifying the availability of paper data, characterized in that: include: The acquisition module is used to obtain the text of the paper to be classified; A processing module, used for performing data processing on the paper text to be classified, and extracting the paper data availability declaration text, wherein the paper data availability declaration text is the text in the paper used to describe the availability of paper data; A classification module is used to use the data availability declaration text of the paper as the input of a data availability classification model to obtain an availability classification result; wherein the data availability classification model is used to preprocess the data availability declaration text of the paper and convert it into a vector sequence, and perform a convolution operation on the vector sequence to obtain text features of the data availability declaration text of the paper, and map the text features to a preset classification collection to obtain the availability classification result of the data availability declaration text of the paper.

9. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for classifying the availability of paper data as described in any one of claims 1 to 7 is implemented.

10. A computer storage medium, characterized in that: The computer storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the paper data availability classification method as described in any one of claims 1 to 7.

Images