t-sne-based high-entropy alloy document keyword implicit semantic mining method and device

Abstract

The application provides a t-SNE-based high-entropy alloy document keyword implicit semantic mining method and device, and relates to the technical field of material document clustering. The method comprises the following steps: obtaining and preprocessing an entropy alloy document data set; training a high-entropy alloy document word vector; representing a high-entropy alloy document; clustering analyzing a high-entropy alloy document vector; and mining a high-entropy alloy hotspot and development trend according to a clustering analysis result. The high-entropy alloy clustering method comprehensively considers the abstract and keyword information of the high-entropy alloy document, fully mines the implicit semantic information from the high-entropy alloy abstract, fully mines the implicit semantic information of the high-entropy alloy document keyword, and through the t-SNE visualized high-entropy alloy document clustering, the distribution and clustering effect of the high-entropy alloy document can be more intuitively observed, thereby providing technical support for mining potential information such as the development trend of the high-entropy alloy.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of material document clustering analysis, and in particular to a t-SNE high-entropy alloy document keyword implicit semantic mining method and device. BACKGROUND

[0002] High-entropy alloys have more excellent mechanical, physical and chemical properties than traditional alloys, have extremely wide application prospects, and are a major hotspot in the field of material science. In order to quickly lock the hot research direction of high-entropy alloys, it is necessary to statistically analyze the vast amount of high-entropy alloy documents.

[0003] In recent years, a lot of achievements have been made in the research on text clustering in the field of data mining. Many methods are based on representing documents as vectors, and clustering algorithms are used to analyze the clustering of documents. The high-entropy alloy document contains a large amount of non-structured information, so clustering can be applied to high-entropy alloy document analysis. At present, there have been clustering researches on high-entropy alloy documents, such as high-entropy alloy document clustering analysis based on co-occurrence network and high-entropy alloy document clustering analysis based on citation network. It can be seen that the current high-entropy alloy document clustering analysis is mainly based on bibliometrics research, and the semantic information is not considered. SUMMARY

[0004] The application provides a t-SNE-based high-entropy alloy document keyword implicit semantic mining method and device, which solves the problem that the current high-entropy alloy document clustering analysis is mainly based on bibliometrics research and does not consider semantic information.

[0005] To solve the above-mentioned application purposes, the technical scheme provided by the application is as follows: a t-SNE-based high-entropy alloy document keyword implicit semantic mining method, characterized by the following steps:

[0006] S1: obtaining high-entropy alloy documents in a database; extracting abstracts, keywords and publication time data in the high-entropy alloy documents, pre-processing the abstract and keyword data to obtain processed data;

[0007] S2: obtaining the processed data, taking the abstract part of the high-entropy alloy document as a corpus, performing word vector training through a word vector model to obtain a word vector training result;

[0008] S3: obtaining the word vector training result, judging the frequency of the keywords of the high-entropy alloy document appearing in the abstract; taking the frequency as a weight, representing the high-entropy alloy document through a word frequency weighted keyword vector to obtain an analysis vector of the high-entropy alloy document;

[0009] S4: Project the analysis vector to a two-dimensional plane by t-SNE to obtain the distribution characteristics of high-entropy alloy literature, and select a clustering algorithm according to the distribution characteristics to cluster the high-entropy alloy literature;

[0010] S5: According to the clustering results, determine the theme of each class of high-entropy alloy literature research through keywords, compare the differences and connections between different themes, mine the hotspots and development trends of high-entropy alloys, and complete the t-SNE-based high-entropy alloy literature implicit semantic mining.

[0011] Preferably, in step S1, the abstract and keyword data are preprocessed to obtain processed data, including:

[0012] The abstract and keywords of the high-entropy alloy literature are preprocessed respectively, and special characters and punctuation marks in the high-entropy alloy literature that do not contain any information are deleted; characters with the same meaning are unified in format; chemical formula characters unique to the high-entropy alloy field are preprocessed to obtain processed data.

[0013] Preferably, in step S2, the word vector is trained by the word vector model to obtain a word vector training result, including:

[0014] The word vector is trained by the word vector model, and the trained word vector is combined by a plurality of vector operations to obtain a vector consistent with chemical intuition; wherein the plurality of vector operations include projection, addition and subtraction.

[0015] Preferably, step S2 further comprises:

[0016] The accuracy of the obtained vector is judged according to the following formula (1):

[0017]

[0018] Wherein, P represents the accuracy; N represents the total number of class tasks; T represents the correct class task.

[0019] Preferably, in step S3, the high-entropy alloy literature is represented according to the following formula (2):

[0020]

[0021] Wherein, X represents the high-entropy alloy literature; w i represents the probability of occurrence of the keyword in all abstracts, x i represents the vector of the keyword.

[0022] Preferably, in step S4, the distribution characteristics of the high-entropy alloy literature include:

[0023] The data points will keep the distribution of the analysis vector, that is, two originally adjacent points will still be adjacent, and two points far apart will still be relatively far apart.

[0024] Preferably, in step S4, a clustering algorithm is selected according to the distribution characteristics to cluster the high-entropy alloy literature, including:

[0025] According to the distribution characteristics of the high-entropy alloy literature, a suitable clustering algorithm is selected to cluster the high-entropy alloy literature, the clustering results are inspected and evaluated by t-SNE, the text clustering link is influenced by dimension filtering and cluster granularity setting, and the clustering is adjusted and optimized.

[0026] A high-entropy alloy literature keyword implicit semantic mining device based on t-SNE, the device is used for the high-entropy alloy literature keyword implicit semantic mining method based on t-SNE, the device includes:

[0027] The preprocessing module is used for obtaining high-entropy alloy literature in the database; extracting the abstract, keywords and publication time data in the high-entropy alloy literature, preprocessing the abstract and keyword data to obtain processed data;

[0028] The vector training module is used for obtaining the processed data, taking the abstract part of the high-entropy alloy literature as a corpus, performing word vector training through a word vector model, and obtaining a word vector training result;

[0029] The weight generation module is used for obtaining the word vector training result, judging the frequency of the keywords of the high-entropy alloy literature appearing in the abstract; taking the frequency as the weight, representing the high-entropy alloy literature through the word frequency weighted keyword vector, and obtaining the analysis vector of the high-entropy alloy literature;

[0030] The clustering module is used for projecting the analysis vector to a two-dimensional plane through t-SNE, obtaining the distribution characteristics of the high-entropy alloy literature, and selecting a clustering algorithm according to the distribution characteristics to cluster the high-entropy alloy literature;

[0031] The implicit semantic mining module is used for determining the theme of each class of high-entropy alloy literature research according to the clustering results through keywords, comparing the differences and connections between different themes, mining the hotspots and development trends of high-entropy alloys, and completing the implicit semantic mining of high-entropy alloy literature based on t-SNE.

[0032] Preferably, the preprocessing module is further used for preprocessing the abstract and keywords of the high-entropy alloy literature respectively, deleting special characters and punctuation marks in the high-entropy alloy literature that do not contain any information; unifying the format of characters with the same meaning; preprocessing the chemical formula characters unique to the high-entropy alloy field to obtain the processed data.

[0033] Preferably, the vector training module is further configured to train the word vector by using the word vector model, and combine the trained word vector with a plurality of vector operations to obtain a vector consistent with chemical intuition; wherein the plurality of vector operations include projection, addition and subtraction.

[0034] In an aspect, an electronic device is provided, which includes a processor and a memory having stored therein at least one instruction which is loaded and executed by the processor to implement the above-mentioned t-SNE-based high-entropy alloy document keyword latent semantic mining method.

[0035] In an aspect, a computer-readable storage medium is provided, which has stored therein at least one instruction which is loaded and executed by a processor to implement the above-mentioned t-SNE-based high-entropy alloy document keyword latent semantic mining method. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative effort based on these drawings.

[0037] Figure 1 is a flow diagram of the t-SNE-based high-entropy alloy document keyword latent semantic mining method provided by the embodiments of the present application;

[0038] Figure 2 is a word vector diagram of metals and metal oxides provided by the embodiments of the present application;

[0039] Figure 3 is a document clustering result diagram of high-entropy alloys provided by the embodiments of the present application;

[0040] Figure 4 is a distribution diagram of the number of publications of high-entropy alloy hot direction documents from 1996 to 2022 provided by the embodiments of the present application;

[0041] Figure 5 is a block diagram of the t-SNE-based high-entropy alloy document keyword latent semantic mining device provided by the embodiments of the present application;

[0042] Figure 6 is a structural schematic diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0043] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without any creative effort belong to the scope of protection of the present application.

[0044] The present application aims at the problem that the existing high-entropy alloy literature clustering analysis is mainly based on bibliometrics research and does not consider semantic information, and provides a high-entropy alloy literature keyword implicit semantic mining method based on t-SNE.

[0045] As shown in Figure 1 The embodiments of the present application provide a high-entropy alloy literature keyword implicit semantic mining method based on t-SNE, which can be realized by an electronic device. As shown in Figure 1 The flowchart of the high-entropy alloy literature keyword implicit semantic mining method based on t-SNE, the processing flow of the method can include the following steps:

[0046] S101: Obtain high-entropy alloy literature in a database; extract abstract, keyword and publication time data in the high-entropy alloy literature, pre-process the abstract and keyword data to obtain processed data;

[0047] In a feasible implementation, the abstract and the keyword of the high-entropy alloy literature are pre-processed respectively, and special characters and punctuation marks in the high-entropy alloy literature that do not contain any information are deleted; characters with the same meaning are uniformly formatted; chemical formula characters unique to the high-entropy alloy field are pre-processed to obtain processed data.

[0048] In a feasible implementation, the literature papers related to high-entropy alloys are obtained by logging in the Web of Science database and the like, the abstract, keyword, publication time and other information of the literature are obtained, and the information is saved in an excel table. The abstract and the keyword of the high-entropy alloy literature are pre-processed respectively, and special characters and punctuation marks in the text that do not contain any information are deleted; characters with the same meaning are uniformly formatted; and general pre-processing operations such as the above are needed, and the characteristics unique to the high-entropy alloy field need to be pre-processed, such as the processing of chemical formula characters.

[0049] In the embodiment of the application, English literature in the field of high-entropy alloys is selected as a data source, the core data set of Web of Science is selected, "High Entropy Alloy" is selected as a related theme, the time span is 1996-2022, the literature type is "paper", 13055 pieces of literature in the field of high-entropy alloys are retrieved, and the title, author, abstract, keyword, publication year and other information of the high-entropy alloy literature are extracted.

[0050] In a feasible implementation manner, the high-entropy alloy literature data preprocessing mainly includes text preprocessing of the abstract and the keyword. The abstract preprocessing is for the subsequent word vector training and weight calculation, and the keyword preprocessing is for more accurate representation of the high-entropy alloy literature and literature clustering and reduction of the influence of noise.

[0051] The processing of the abstract includes:

[0052] a. The leading or trailing copyright information and punctuation marks (",", ":", "." and the like) are removed through a regular expression;

[0053] b. The chemical entities and abbreviations are marked through Chem Data Extractor 5, and the chemical entities and abbreviations are kept unchanged;

[0054] c. The common words are converted into lowercase form, and the form is unified.

[0055] The processing of the keyword includes:

[0056] I. Phrases such as High entropy alloys and high-entropy alloy indicating high-entropy alloys are uniformly written as HEA;

[0057] II. Phrases such as Wavelength Dispersive Spectrometry (WDS) having both full name and abbreviation are kept in the form of abbreviation WDS to avoid repetition of information;

[0058] III. Selective lowercase of words, the chemical expression words are kept in the original form, and the common meaning words are processed in lowercase;

[0059] IV. Remove meaningless prepositions and virtual words such as "of", "the", "by" and "to".

[0060] S102: Obtain the processed data, take the abstract part of the high-entropy alloy literature as a corpus, perform word vector training through a word vector model, and obtain a word vector training result;

[0061] In an implementable embodiment, word vectors are trained by using a word vector model, and the trained word vectors are combined by using a plurality of vector operations to obtain vectors consistent with chemical intuition; wherein the plurality of vector operations include projection, addition and subtraction.

[0062] In an implementable embodiment, step S102 further includes:

[0063] The obtained vectors are subjected to accuracy judgment according to formula (1) as follows:

[0064]

[0065] wherein P represents accuracy, N represents the total number of category tasks, and T represents a category task determined to be correct.

[0066] In an implementable embodiment, the accuracy of the category task is used to evaluate the excellence of the word vector training, for example, Zr-ZrO2≈Cr-Cr2O3≈Ni-NiO, and the most similar word of Zr-ZrO2+Cr2O3 is Cr, which is determined to be correct, and the most similar word of Zr-ZrO2+NiO is not Ni, which is determined to be incorrect; therefore, the accuracy is determined according to formula (2).

[0067] In the embodiment of the application, the word vector can be used to represent the semantic relationship of the word, the keywords of the literature are the key information and knowledge points of the literature, and the word vector of the keyword can be applied to the high-entropy alloy literature representation, the method can integrate the semantic information of the keyword into the high-entropy alloy literature vector, and the information contained in the high-entropy alloy literature vector can be better improved.

[0068] S103: Obtain a word vector training result, judge the frequency of the keywords of the high-entropy alloy literature appearing in the abstract, take the frequency as a weight, represent the high-entropy alloy literature by using the word frequency weighted keyword vector, and obtain an analysis vector of the high-entropy alloy literature.

[0069] In an implementable embodiment, the high-entropy alloy literature is represented according to formula (2) as follows:

[0070]

[0071] wherein X represents the high-entropy alloy literature, w i represents the appearance probability of the keyword in all abstracts, x i represents the vector of the keyword.

[0072] In a feasible implementation, the abstract part (preprocessed) of 13055 high-entropy alloy documents is used as a corpus to train word vectors by using the Skip-gram model of Word2Vec. The window size is set to 5, and the dimension of the word vector is set to 200. The word vector is trained through negative sampling optimization. The accuracy in 3000 high-entropy alloy documents and 3000 syntactic analogy tasks is as shown in Table 1.

[0073] Table 1 Accuracy of word vector category task

[0074] High-entropy alloy Grammar Average 50.2 52.6 51.4

[0075] The frequency of each keyword is counted according to the abstract information, and the keyword vector is represented by the keyword frequency weighting.

[0076] S104: The analysis vector is projected to a two-dimensional plane by t-SNE to obtain the distribution characteristics of the high-entropy alloy documents, and a clustering algorithm is selected according to the distribution characteristics to cluster the high-entropy alloy documents.

[0077] In a feasible implementation, the distribution characteristics of the high-entropy alloy documents include:

[0078] The data points will maintain the same distribution, that is, two originally adjacent points will still be adjacent, and two points far away from each other will still be relatively far away.

[0079] In a feasible implementation, the high-entropy alloy documents are clustered according to the distribution characteristics by selecting a clustering algorithm, including:

[0080] According to the distribution characteristics of the high-entropy alloy documents, a suitable clustering algorithm is selected to cluster the high-entropy alloy documents, the clustering results are observed and evaluated by t-SNE, the text clustering link is affected by dimension screening and clustering granularity setting, and the clustering is adjusted and optimized.

[0081] In the embodiment of the application, the document vector of the high-entropy alloy belongs to a high-dimensional vector, and the distribution characteristics of the high-entropy alloy documents cannot be directly observed, which will cause unnecessary trouble for the selection of the clustering algorithm and data analysis. The high-entropy alloy vector can be projected to a low-dimensional space for visualization by t-SNE.

[0082] In a feasible implementation, after the word vector is projected to a two-dimensional space by t-SNE, there is a consistent vector operation between the metal and its corresponding oxide, as shown in the accompanying Figure 2

[0083] ​In one feasible implementation, the vectors of high-entropy alloy literature are projected onto a two-dimensional plane. Based on the circular distribution of these literatures, hierarchical clustering is used to cluster them. The Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm was used to cluster 13055 high-entropy alloy literatures. The maximum sample radius threshold T for the clustering feature (CF) of leaf nodes was set to 0.67, and the maximum CF number for both internal nodes and leaf nodes was set to 50. The clustering results are shown in the attached figure. Figure 3 As shown in Table 2, the number of high-entropy alloy publications and the frequency of keywords in each category were statistically analyzed. The top 10 most frequent keywords for each category were listed, and the themes of that category were analyzed accordingly. The table also shows the annual publication statistics for high-entropy alloys in each category from 1996 to 2022. Figure 3 As shown.

[0084] Table 2 Cluster Analysis of High-Entropy Alloys at Different Periods

[0085]

[0086]

[0087] S105: Based on the clustering results, determine the research topics of each type of high-entropy alloy literature through keywords, compare the differences and connections between different topics, explore the hot spots and development trends of high-entropy alloys, and complete the implicit semantic mining of high-entropy alloy literature based on t-SNE.

[0088] In one feasible implementation, from the attached Figure 3 As shown in Table 2, the literature on high-entropy alloys can be broadly categorized into seven types. Among these, "mechanical" and "microstructure" are addressed in six categories (A, B, C, E, F, and G), indicating that mechanical properties and microstructure are the main research directions for high-entropy alloys. In addition, other hot research areas for high-entropy alloys include:

[0089] (1) Category A includes research on the magnetic and thermodynamic aspects of high-entropy alloys, represented by keywords such as "magnetic", "thermodynamic", and "thermal".

[0090] (2) Category B is represented by keywords such as "high", "medium", "entropy" and "alloy", which are related studies on high-entropy alloys and medium-entropy alloys;

[0091] (3) In the C category, the keywords such as "laser", "cladding", "corrosion", "wear", "hardness" are representative, and the researches on the application of laser cladding in high-entropy alloys and the corrosion resistance, wear resistance and high hardness characteristics of high-entropy alloys;

[0092] (4) In the D category, the keywords such as "laser", "diffusion", "twinning", "grain", "hydrogen" are representative, and the researches on the application of laser technology in high-entropy alloys, the diffusion and twinning of high-entropy alloys and the performance related aspects of hydrogen storage;

[0093] (5) In the E category, the keywords such as "high", "temperature", "oxidation", "pressure" are representative, and the researches on the characteristics of high-entropy alloys under high temperature and high pressure and oxidation resistance;

[0094] (6) In the F category, the keywords such as "phase", "transformation", "stability", "transition" are representative, and the researches on the phase change, phase stability and phase transition of high-entropy alloys;

[0095] (7) In the G category, the keywords such as "deformation", "strengthening", "strain" are representative, and the researches on the deformation mechanism and strengthening related aspects of high-entropy alloys.

[0096] By the attached Figure 4As can be seen from the figure, during 1996-2022, the research of high-entropy alloy hot spot direction is in an upward state. During 1996-2000, the research of high-entropy alloy is less; during 2001-2008, only the research of hot spot A is relatively more, and the research of other hot spots is still in a low state; during 2009-2015, more hot spot research has a slow rise, such as hot spot B and F, and the high-entropy alloy related literature published in hot spot A and B research direction reached 100 in 2010; during 2016-2018, each hot spot research has a faster rise, and the high-entropy alloy related literature published in 2018 basically reached 100, and hot spot B has basically reached the research heat of hot spot A; during 2019-2022, the research of high-entropy alloy enters a rapid rising period, and the number of high-entropy alloy literature published in each hot spot field has greatly improved, among which the number of high-entropy alloy literature published in hot spots B, C and D exceeds 300 in 2021, respectively ranks the first, the second and the third, and the rising trend has no signs of weakening in the next two years.

[0097] In summary, the research of high-entropy alloy has been in an increasing state, and in the past 5 years, it has been a period of rapid increase in the research of high-entropy alloy, and the research of high-entropy alloy will continue to increase in the future 2 years. It can be predicted that in the next 2 years, the research direction of high-entropy alloy will still be mechanical properties and microstructure. Hot spots B, C and D will attract more attention than other hot spot research, among which, medium-entropy alloy as the main part of B hot spot research content may become another major material research hot spot after high-entropy alloy; laser technology plays an important role in C and D hot spot research, and will play a greater role in high-entropy alloy in the future; in addition, the research on the corrosion resistance, wear resistance and high hardness characteristics of high-entropy alloy in hot spot C and the research on the diffusion, twinning and hydrogen storage related aspects of high-entropy alloy in hot spot D will also be the key research direction.

[0098] In the embodiment of the application, similar theme high-entropy alloy literature is clustered into a category by literature clustering, the research theme of each category of high-entropy alloy literature is analyzed, and the development trend of the high-entropy alloy field is predicted, so that the research work of high-entropy alloy is carried out more scientifically. The high-entropy alloy clustering method comprehensively considers the abstract and keyword information of the high-entropy alloy literature, fully excavates the implicit semantic information from the high-entropy alloy abstract, uses the word vector to extract the semantic information of the keyword, extracts the weight information from the abstract, and performs distributed representation on the high-entropy alloy literature in a weighted manner through the word vector, so that the high-entropy alloy literature can be better represented; the high-entropy alloy literature is projected to a two-dimensional plane through t-SNE, so that the distribution and clustering effect of the high-entropy alloy literature can be more intuitively observed, and technical support is provided for mining potential information such as the development trend of high-entropy alloy.

[0099] In the embodiment of the present application, the English literature in the field of high-entropy alloys is clustered and analyzed by Word2Vec and BIRCH clustering algorithm. It can be predicted that the research interest of high-entropy alloys will continue to rise in the next two years, the research content of high-entropy alloys will continue to focus on microstructure and mechanical properties, laser technology will be applied more and more in high-entropy alloys, and the research on the corrosion resistance, wear resistance and high hardness of high-entropy alloys will continue to deepen. It is indicated that the research topics in the future are the design of high-entropy alloys for performance (demand) and the optimization and exploration of high-entropy alloy synthesis and processing technology. The present study applies machine learning to statistically quantify the literature in the field of high-entropy alloys, which has a guiding effect on the research in this field.

[0100] Figure 5 is a t-SNE-based high-entropy alloy literature keyword implicit semantic mining device of the present application, the device 200 is used for the above-mentioned t-SNE-based high-entropy alloy literature keyword implicit semantic mining, the device 200 comprises:

[0101] A preprocessing module 210 is configured to obtain high-entropy alloy literature in a database, extract abstracts, keywords and publication time data from the high-entropy alloy literature, and preprocess the abstracts and keywords data to obtain processed data.

[0102] A vector training module 220 is configured to obtain the processed data, take the abstracts of the high-entropy alloy literature as a corpus, perform word vector training through a word vector model, and obtain a word vector training result.

[0103] A weight generation module 230 is configured to obtain the word vector training result, judge the frequency of the keywords of the high-entropy alloy literature appearing in the abstracts, take the frequency as a weight, represent the high-entropy alloy literature through a word frequency weighted keyword vector, and obtain an analysis vector of the high-entropy alloy literature.

[0104] A clustering module 240 is configured to project the analysis vector to a two-dimensional plane through t-SNE, obtain the distribution characteristics of the high-entropy alloy literature, and select a clustering algorithm according to the distribution characteristics to cluster the high-entropy alloy literature.

[0105] An implicit semantic mining module 250 is configured to determine the theme of the research of each type of high-entropy alloy literature through keywords according to the clustering result, compare the differences and connections between different themes, mine the hotspots and development trends of high-entropy alloys, and complete the t-SNE-based implicit semantic mining of high-entropy alloy literature.

[0106] Preferably, the preprocessing module 210 is further configured to preprocess the abstracts and keywords of the high-entropy alloy documents respectively, delete special characters and punctuation marks in the high-entropy alloy documents that do not contain any information, unify the format of characters with the same meaning, and preprocess chemical formula characters unique to the high-entropy alloy field to obtain processed data.

[0107] Preferably, the vector training module 220 is further configured to train word vectors by using a word vector model, combine the trained word vectors through a plurality of vector operations, and obtain vectors consistent with chemical intuition; wherein the plurality of vector operations include projection, addition, and subtraction.

[0108] Preferably, the vector training module 220 is further configured to judge the accuracy of the obtained vectors according to the following formula (1):

[0109]

[0110] wherein P represents accuracy, N represents the total number of category tasks, and T represents a category task determined to be correct.

[0111] Preferably, the weight generation module 230 is further configured to represent the high-entropy alloy document according to the following formula (2):

[0112]

[0113] wherein X represents the high-entropy alloy document, w i represents the probability of occurrence of the keyword in all abstracts, x i represents the vector of the keyword.

[0114] Preferably, the clustering module 240 is further configured to keep the distribution of data points and analysis vectors, that is, two originally adjacent points will still be adjacent, and two points far apart will still be relatively far apart.

[0115] Preferably, the clustering module 240 is further configured to select a suitable clustering algorithm for clustering the high-entropy alloy documents according to the distribution characteristics of the high-entropy alloy documents, use t-SNE to understand and evaluate the clustering results, and use dimension screening and cluster granularity setting to influence the text clustering link and adjust and optimize the clustering.

[0116] In the embodiment of the present application, similar high-entropy alloy literature is clustered into a category through literature clustering, the research topic of each category of high-entropy alloy literature is analyzed, and the development trend of the high-entropy alloy field is predicted, so that the research work of high-entropy alloys can be carried out more scientifically. The high-entropy alloy clustering method comprehensively considers the abstract and keyword information of high-entropy alloy literature, fully excavates the implicit semantic information from the abstract of high-entropy alloy, uses the word vector to extract the semantic information of the keyword, extracts the weight information from the abstract, and performs distributed representation on the high-entropy alloy literature through the word vector weighting method, which can better represent the high-entropy alloy literature; the high-entropy alloy literature is projected to a two-dimensional plane through t-SNE, which can more intuitively observe the distribution and clustering effect of high-entropy alloy literature, and provide technical support for mining potential information such as the development trend of high-entropy alloys.

[0117] Figure 6 Figure 1 is a structural schematic diagram of an electronic device 300 provided by the embodiment of the present application. The electronic device 300 can have great differences due to different configurations or performances, and can include one or more processors (central processing units, CPUs) 301 and one or more memories 302. The memory 302 stores at least one instruction, which is loaded and executed by the processor 301 to implement the steps of the following t-SNE-based high-entropy alloy literature implicit semantic mining method:

[0118] S1: Obtain high-entropy alloy literature in a database; extract the abstract, keywords and publication time data in the high-entropy alloy literature, and pre-process the abstract and keyword data to obtain processed data;

[0119] S2: Obtain the processed data, take the abstract part of the high-entropy alloy literature as a corpus, perform word vector training through a word vector model, and obtain a word vector training result;

[0120] S3: Obtain the word vector training result, judge the frequency of the keywords of the high-entropy alloy literature appearing in the abstract; take the frequency as a weight, represent the high-entropy alloy literature through a word frequency weighted keyword vector, and obtain an analysis vector of the high-entropy alloy literature;

[0121] S4: Project the analysis vector to a two-dimensional plane through t-SNE, obtain the distribution characteristics of the high-entropy alloy literature, and select a clustering algorithm according to the distribution characteristics to cluster the high-entropy alloy literature;

[0122] S5: According to the clustering result, determine the research topic of each category of high-entropy alloy literature through keywords, compare the differences and connections between different topics, mine the hotspots and development trends of high-entropy alloys, and complete the t-SNE-based high-entropy alloy literature implicit semantic mining.

[0123] In an exemplary embodiment, a computer readable storage medium, such as a memory including instructions executable by a processor in a terminal to perform the above-described t-SNE-based high-entropy alloy document latent semantic mining method, is also provided. For example, the computer readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0124] Those of ordinary skill in the art understand that all or part of the steps of the above-described embodiments can be completed by hardware or by programs instructing relevant hardware, and the programs can be stored in a computer readable storage medium, such as a read-only memory, a magnetic disk, or an optical disk.

Claims

1. A t-SNE-based high-entropy alloy document implicit semantic mining method, characterized by the following steps The method comprises the following steps: S1: obtaining high-entropy alloy literature in a database; extracting abstract, keywords and publication time data in the high-entropy alloy literature, preprocessing the abstract and keyword data to obtain processed data; S2: obtaining the processed data, taking the abstract part of the high-entropy alloy literature as a corpus, performing word vector training through a word vector model to obtain a word vector training result; S3: obtaining the word vector training result, judging the frequency of the keywords of the high-entropy alloy literature appearing in the abstract; taking the frequency as a weight, representing the high-entropy alloy literature through a keyword frequency weighted keyword vector to obtain an analysis vector of the high-entropy alloy literature; S4: projecting the analysis vector to a two-dimensional plane through t-SNE to obtain the distribution characteristics of the high-entropy alloy literature, and selecting a clustering algorithm according to the distribution characteristics to cluster the high-entropy alloy literature; S5: according to the clustering result, determining the theme of each type of high-entropy alloy literature research through keywords, comparing the differences and connections between different themes, mining the hot spots and development trends of high-entropy alloys, and completing the t-SNE-based high-entropy alloy literature implicit semantic mining.

2. The method of claim 1, wherein, In step S1, the abstract and keyword data are preprocessed to obtain processed data, which comprises: The abstract and keywords of the high-entropy alloy literature are preprocessed respectively, and special characters and punctuation marks in the high-entropy alloy literature that do not contain any information are deleted; characters with the same meaning are uniformly formatted; chemical formula characters unique to the high-entropy alloy field are preprocessed to obtain processed data.

3. The method of claim 1, wherein, In step S2, the word vector training result is obtained by training the word vector through a word vector model, which comprises: The word vector is trained by using a word vector model, and the trained word vector is combined through a plurality of vector operations to obtain a vector consistent with chemical intuition; wherein the plurality of vector operations include projection, addition and subtraction.

4. The method of claim 3, wherein, The step S2 further comprises: The accuracy of the obtained vector is judged according to the following formula (1): Wherein, P represents the accuracy; N represents the total number of category tasks; T represents the category task judged correctly.

5. The method of claim 1, wherein, In step S3, the high-entropy alloy literature is represented according to the following formula (2): where X represents high-entropy alloy literature; w i represents the probability of the keyword appearing in all abstracts, x i represents the keyword vector.

6. The method of claim 5, wherein, In step S4, the distribution characteristics of the high-entropy alloy literature comprise: The data points will maintain the same distribution as the analysis vector, that is, two originally adjacent points will still be adjacent, and two points far apart will still be relatively far apart.

7. The method of claim 5, wherein, In step S4, the high-entropy alloy literature is clustered according to the distribution characteristics, which comprises: According to the distribution characteristics of the high-entropy alloy literature, a suitable clustering algorithm is selected to cluster the high-entropy alloy literature, the clustering result is understood and evaluated through t-SNE, the dimension is screened, the clustering granularity is set to affect the text clustering link, and the clustering is adjusted and optimized. 8.A device for t-SNE-based high-entropy alloy literature keyword latent semantic mining, characterized in that, The device is used for the t-SNE-based high-entropy alloy literature keyword implicit semantic mining method according to any one of claims 1-7, and the device comprises: The preprocessing module is configured to obtain high-entropy alloy literatures in a database, extract abstracts, keywords and publication time data in the high-entropy alloy literatures, and pre-process the abstracts and keywords to obtain processed data. The vector training module is configured to obtain the processed data, take the abstracts of the high-entropy alloy literatures as a corpus, perform word vector training through a word vector model, and obtain a word vector training result. The weight generation module is configured to obtain the word vector training result, determine a frequency of occurrence of the keywords in the abstracts of the high-entropy alloy literatures, take the frequency as a weight, represent the high-entropy alloy literatures through a word frequency weighted keyword vector, and obtain an analysis vector of the high-entropy alloy literatures. The clustering module is configured to project the analysis vector to a two-dimensional plane through t-SNE, obtain distribution characteristics of the high-entropy alloy literatures, select a clustering algorithm according to the distribution characteristics, and cluster the high-entropy alloy literatures. The implicit semantic mining module is configured to determine a theme of research of each type of high-entropy alloy literatures through keywords according to a clustering result, compare differences and connections between different themes, mine hot spots and development trends of high-entropy alloys, and complete implicit semantic mining of the high-entropy alloy literatures based on t-SNE.

9. The apparatus of claim 8, wherein, The preprocessing module is further configured to pre-process the abstracts and keywords of the high-entropy alloy literatures respectively, delete special characters and punctuation symbols in the high-entropy alloy literatures that do not contain any information, unify formats of characters with the same meaning, pre-process chemical formula characters unique to the high-entropy alloy field, and obtain processed data.

10. The apparatus of claim 9, wherein, The vector training module is further configured to train word vectors through a word vector model, perform a plurality of vector operation combinations on the trained word vectors, and obtain vectors consistent with chemical intuition; the plurality of vector operations include projection, addition and subtraction.

Images