Scientific research achievement person name disambiguation and achievement merging method, system, terminal and medium

Abstract

The application provides a scientific research achievement person name disambiguation and achievement merging method, system, terminal and medium, which is not only suitable for the field of scientific research, but also can be extended to other fields, and provides an effective solution for solving the same name problem in various fields; through accurate person name disambiguation and achievement merging, the literature reference and retrieval are more accurate and efficient, and more convenient and accurate literature information is provided for researchers; through merging the achievements of the same name authors, a more complete and accurate scientific knowledge graph can be constructed, and more rich information and more accurate correlation are provided for academic research and subject development; various information such as text features, cooperation relationship, institution information and reference relationship are combined, so that the disambiguation and merging are more comprehensive and integrated, the accuracy and robustness of the method are improved; the identity of the same name scientific research author can be effectively identified, the scientific research achievements of different authors are avoided from being incorrectly attributed to the same author, so that the accuracy and reliability of the scientific research achievements are ensured.

Description

Technical Field

[0001] This application relates to the field of computer natural language processing technology, and in particular to methods, systems, terminals and media for disambiguation of names and merging of research results. Background Technology

[0002] Currently, the goal of person name disambiguation is to accurately match individuals with the same name to their corresponding real identities, thereby eliminating confusion and ensuring the accuracy and reliability of the data. Person name disambiguation has important applications in many fields, with scientific research being particularly prominent. In scientific research, academic papers, research results, patents, and other documents frequently involve the names of scientists, scholars, and authors. Due to the existence of authors with the same name, the names in these documents may correspond to different individuals; therefore, person name disambiguation is needed to accurately associate these names with the correct individuals.

[0003] The challenge of name disambiguation lies in the lack of explicit contextual or correlative information, making it difficult to accurately distinguish individuals with the same name. To address this, researchers typically utilize other features in the textual data, such as the author's affiliation, published paper titles, and collaboration relationships, to assist in name disambiguation. In academic research, accurately identifying authors with the same name ensures that research findings are correctly cited and attributed, protects academic reputation, and promotes academic exchange and collaboration. While name disambiguation can be achieved to some extent using simple methods such as rules and normalized names, the effectiveness of traditional methods may become limited as the scale and complexity of the data increase. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method, system, terminal and medium for disambiguation of scientific research results and merging of results, which is used to solve the problems of low accuracy and poor effect of disambiguation of names in the prior art.

[0005] To achieve the above and other related objectives, a first aspect of this application provides a method for disambiguating names and merging research results, comprising: obtaining a dataset of research results containing names to be disambiguated; the dataset includes at least a dataset of papers and a dataset of patents; preprocessing and partitioning the dataset based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name; extracting the paper feature vectors of the papers in the set of papers with the same name based on the SPECTER model; extracting the patent feature vectors of the patents in the set of patents with the same name based on the TF-IDF model; calculating the paper feature vectors and patent feature vectors based on a similarity algorithm and a rule-based model, respectively, to obtain a paper similarity matrix and a patent similarity matrix; processing the paper similarity matrix based on a nearest neighbor propagation algorithm and a HAC clustering algorithm to obtain a paper disambiguation result; and processing the patent similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain a patent disambiguation result; and merging the paper dataset in the paper disambiguation result with the patent dataset of the corresponding scholar in the patent disambiguation result to obtain a dataset of research results for each scholar.

[0006] In some embodiments of the first aspect of this application, the step of preprocessing and partitioning the research results dataset based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name specifically includes: standardizing the names of scholars in the research results dataset based on the LN-FI criterion to obtain a unified scholar name format; and partitioning the research results dataset according to the unified scholar name format to obtain a set of papers with the same name and a set of patents with the same name for each scholar with the same name in the research results dataset.

[0007] In some embodiments of the first aspect of this application, the SPECTER model is calculated as follows: E = M([T1,T2,…,T…) i ]); where T i This represents the text information of a certain field, M represents the SPECTER model, and E is the output feature vector.

[0008] In some embodiments of the first aspect of this application, the calculation process of the TF-IDF model includes the following:

[0009]

[0010] Where j represents the article, i represents the words in the article, k refers to the total number of words in article j, and tf i,j |D| represents the frequency of each word in the article, |D| represents the total number of articles in the corpus, and |{j:t i ∈d j | represents the number of articles in the collection that contain this word, idf iIndicates the inverse document frequency of word i.

[0011] In some embodiments of the first aspect of this application, the calculation of the paper feature vector and the patent feature vector using the similarity algorithm and the rule-based model respectively to obtain the paper similarity matrix and the patent similarity matrix specifically includes: using the similarity algorithm to perform semantic similarity calculation on the paper feature vector and the patent feature vector respectively to obtain a first similarity matrix of the paper feature vector and a first similarity matrix of the patent feature vector; using the rule-based model to perform similarity judgment on the paper feature vector and the patent feature vector respectively to obtain a second similarity matrix of the paper feature vector and a second similarity matrix of the patent feature vector; constructing a paper similarity matrix based on the first similarity matrix of the paper feature vector and the second similarity matrix of the paper feature vector; and constructing a patent similarity matrix based on the first similarity matrix of the patent feature vector and the second similarity matrix of the patent feature vector.

[0012] In some embodiments of the first aspect of this application, the calculation process of the HAC clustering algorithm includes the following: treating each data point in the sample set as a cluster; calculating the distance between every two clusters, merging the two closest or most similar clusters into a new cluster; calculating the distance between the new cluster and all other clusters; repeating the above steps until the iteration termination condition is met.

[0013] To achieve the above and other related objectives, a second aspect of this application provides a system for disambiguating and merging names in scientific research achievements, comprising: a scientific research achievement dataset acquisition module for acquiring a dataset of scientific research achievements containing names to be disambiguated; the scientific research achievement dataset includes at least a dataset of papers and a dataset of patents; a preprocessing module for preprocessing and partitioning the scientific research achievement dataset based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name; and a feature vector acquisition module for extracting the paper feature vectors of the papers in the set of papers with the same name based on the SPECTER model; and extracting the patent feature vectors of the patents in the set of patents with the same name based on the TF-IDF model. The system includes: a similarity matrix acquisition module, used to calculate the paper feature vector and patent feature vector based on a similarity algorithm and a rule-based model, respectively, to obtain the paper similarity matrix and the patent similarity matrix; a disambiguation result acquisition module, used to process the paper similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the paper disambiguation result; and to process the patent similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the patent disambiguation result; and a results merging module, used to merge the paper dataset in the paper disambiguation result with the corresponding scholar's patent dataset in the patent disambiguation result to obtain the research results dataset of each scholar.

[0014] In some embodiments of the second aspect of this application, the preprocessing module includes: a standardization processing module, used to standardize the scholar names in the scientific research results dataset based on the LN-FI criterion to obtain a unified scholar name format; and a homonym segmentation module, used to segment the scientific research results dataset according to the unified scholar name format to obtain a set of papers and patents with the same name for each scholar with the same name in the scientific research results dataset.

[0015] To achieve the above and other related objectives, a third aspect of this application provides an electronic terminal, comprising: a processor and a memory; the memory for storing a computer program; and the processor for executing the computer program stored in the memory, so that the electronic terminal executes the method for disambiguating names and merging research results.

[0016] To achieve the above and other related objectives, a fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a method for disambiguating names and merging results in scientific research.

[0017] As described above, the method, system, terminal, and medium for disambiguation of names and merging of research results in this application have the following beneficial effects:

[0018] (1) Wide range of applications: It is not only applicable to the field of scientific research, but can also be extended to other fields, such as medicine and law, providing effective solutions to the same name problem in various fields.

[0019] (2) Improve the efficiency of literature citation and retrieval: Through accurate name disambiguation and results merging, literature citation and retrieval are more accurate and efficient, providing researchers with more convenient and accurate literature information.

[0020] (3) Improve the construction of scientific knowledge graphs: By merging the results of authors with the same name, a more complete and accurate scientific knowledge graph can be constructed, providing richer information and more accurate correlations for academic research and discipline development.

[0021] (4) Introducing multiple feature information: Combining multiple information such as text features, cooperation relationships, institutional information and citation relationships, making disambiguation and merging more comprehensive and integrated, and improving the accuracy and robustness of the method.

[0022] (5) Improve the accurate attribution of scientific research results: It can effectively identify the identity of scientific research authors with the same name, avoid the scientific research results of different authors being wrongly attributed to the same author, thereby ensuring the accuracy and credibility of scientific research results. Attached Figure Description

[0023] Figure 1 The diagram shown is a flowchart illustrating a method for disambiguating names and merging research results in an embodiment of this application.

[0024] Figure 2 The diagram shows a specific embodiment of a method for disambiguating names and merging research results in accordance with this application.

[0025] Figure 3 The diagram shown is a structural schematic of a system for disambiguating names and merging research results in an embodiment of this application.

[0026] Figure 4 The diagram shown is a structural schematic of an electronic terminal according to an embodiment of this application. Detailed Implementation

[0027] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0028] It should be noted that in the following description, reference is made to the accompanying drawings, which illustrate several embodiments of this application. It should be understood that other embodiments may also be used, and changes in mechanical composition, structure, electrical system, and operation may be made without departing from the spirit and scope of this application. The following detailed description should not be considered limiting, and the scope of the embodiments of this application is defined only by the claims of the published patent. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. Spatially related terms, such as “upper,” “lower,” “left,” “right,” “below,” “below,” “lower part,” “above,” “upper part,” etc., may be used herein to illustrate the relationship between one element or feature shown in the figures and another element or feature.

[0029] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," "fixing," and "holding" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0030] Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context indicates otherwise. It should be further understood that the terms “comprising,” “including,” indicate the presence of the stated feature, operation, element, component, item, kind, and / or group, but do not preclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of: A; B; C; A and B; A and C; B and C; A, B, and C.” Exceptions to this definition arise only when combinations of elements, functions, or operations are inherently mutually exclusive in some manner.

[0031] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the invention.

[0032] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0033] <1> Natural Language Processing (NLP) is an important branch of computer science and artificial intelligence that studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is not simply the study of natural language itself, but rather the development of computer systems, especially the software systems, that can effectively achieve natural language communication.

[0034] <2> Name disambiguation: Name disambiguation has important applications in many fields, with scientific research being particularly prominent. In scientific research, academic papers, research results, patents, and other documents frequently involve the names of scientists, scholars, and authors. Due to the existence of authors with the same name, the names in these documents may correspond to different individuals. Therefore, name disambiguation is needed to accurately associate these names with the correct individuals.

[0035] <3> SPECTER: SPECTER stands for Scientific Paper Embeddings using Citation-informed Transformers. It is a pre-trained model for scientific documents proposed by Allen AI in 2020. Its goal is to provide an effective distributed representation method for scientific research, enabling scientific documents to be represented and compared in semantic space, making it very suitable for constructing distributed representations of research results.

[0036] <4> Hierarchical Agglomerative Clustering (HAC) is a highly effective clustering algorithm. Its main idea is to treat each sample point as a cluster and then continuously merge the two closest clusters until a certain iteration termination condition is met. For example, if the current number of clusters is 20% of the initial number, or 80% of the clusters have been merged. In summary, the specific steps of HAC include: treating each data point in the training sample set as a cluster; calculating the distance between every two clusters and merging the two closest or most similar clusters; repeating the above steps until the iteration termination condition is met. In this algorithm, there are four ways to measure similarity: (1) Single-link: the distance between the two closest points in two different clusters, i.e., MIN; (2) Complete-link: the distance between the two farthest points in two different clusters, i.e., MAX; (3) Average-link: the average distance between all point pairs in two different clusters, i.e., AVERAGE; (4) Ward-link: the increment of the sum of squared deviations after two different clusters are clustered together.

[0037] <5> TF-IDF (Term Frequency-Inverse Document Frequency) is a statistical method commonly used in text processing to assess the importance of a word within a document. Simply put, it can be used for extracting keywords from documents. The TF-IDF model is a statistical method and a commonly used weighting technique in information retrieval and data mining to evaluate the importance of a word to a document within a document set or corpus. In recent years, it has been widely applied in scenarios involving the identification, statistics, and filtering of high-frequency keywords, such as user keyword identification and extraction in artificial intelligence scenarios.

[0038] like Figure 1 The diagram illustrates a flowchart of a method for disambiguating names and merging research results according to an embodiment of this application. Figure 2 Instructions. The main steps include the following:

[0039] Step S11: Obtain the research results dataset containing the names to be disambiguated; the research results dataset includes at least a paper dataset and a patent dataset. Obtain the research results dataset for which name disambiguation needs to be performed. The types of research results include papers and patents. Because papers and patents are fundamentally different, the research results dataset is divided into a paper dataset and a patent dataset, so that name disambiguation and result merging can be achieved by operating on the paper dataset and the patent dataset separately.

[0040] Step S12: Preprocess and divide the research results dataset based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name.

[0041] It should be noted that the research data obtained for disambiguating names is mostly in the widely used formats of JSON (JavaScript Object Notation) and XML (eXtensible Markup Language). JSON or XML data typically comes from research data crawled by web crawlers, resulting in a somewhat chaotic data format. For example, there may be inconsistent formats for names with the same surname, and ambiguous names, specifically including surname-first-name, surname-last-name-first, and names with both traditional and simplified Chinese characters. In such cases, the accuracy of name-based segmentation based on this chaotic data format is low. To facilitate subsequent calculations and improve the accuracy of name-based segmentation, preprocessing of the research dataset is necessary. The specific preprocessing process is as follows:

[0042] The names of scholars in the aforementioned research dataset were standardized using the LN-FI criterion to obtain a unified format. The LN-FI criterion, short for Last Name and First Initial, is based on the first letter of the scholar's last name. For example, "Zhang San" is considered "Zhang S.", or "Mary Jane" is considered "Mary J.".

[0043] The research achievement dataset is partitioned according to the standardized scholar name format to obtain sets of papers and patents with the same name for each scholar in the dataset. Specifically, all research achievements in the dataset are divided into different data blocks based on the standardized scholar name format to form sets of papers and patents with the same name for each scholar. For example, partitioning the research achievement dataset for "Zhang San" yields Paper 1, Paper 2, Paper 3, and Paper 4 under the Zhang S. set, and Patent 1, Patent 2, and Patent 3 under the Zhang S. set. Data block partitioning based on the LN-FI criterion is currently a very fine-grained partitioning method for name disambiguation tasks, especially when dealing with inconsistent or ambiguous name formats.

[0044] Step S13: Extract the paper feature vectors of the papers in the same-name paper set based on the SPECTER model; extract the patent feature vectors of the patents in the same-name patent set based on the TF-IDF model.

[0045] It should be noted that inputting the text data of papers from the research dataset into a pre-trained model yields distributed representations, i.e., extracting feature vectors from papers with the same name within a set. Specifically, with the development of natural language processing technology, methods based on deep pre-trained models have achieved outstanding performance on various language tasks. Large-scale pre-training methods can generate rich distributed representations. In the large-scale pre-training stage, the model learns distributed representations of words and text through pre-training on large-scale unlabeled data. These distributed representations contain rich semantic information, making similar words or texts closer together in the vector space.

[0046] The advantage of pre-trained models lies in their ability to encode vast amounts of corpus information into vector representations, thereby providing better feature representations for downstream tasks such as text classification, machine translation, and question answering. The rise and success of large-scale pre-training methods have brought significant breakthroughs to natural language processing, such as BERT (Bidirectional Encoder Representations from Transformers) and the GPT (Generative Pre-trained Transformer) series. These models, through large-scale pre-training, have obtained high-quality distributed representations, greatly promoting the effectiveness and progress of NLP (Natural Language Processing) tasks.

[0047] The distributed representation of a paper means that the text data of the paper includes at least the following features: author information and article information. Author information includes author names, affiliations, and co-author information; article information includes the article title, abstract, conference attended, year of publication, citation information, and information on which the article was cited.

[0048] In this embodiment, the pre-trained models for extracting feature vectors from the papers in the collection of papers with the same name include word2vec, glove, ELMo, BERT, and SPECTER models. SPECTER is preferred in this embodiment. The SPECTER model can model textual information such as article titles and abstracts. Simply concatenate the text information and input it into the model to obtain its distributed representation; that is, the output distributed representation is the feature vector of the paper. For example, the distributed representation output for the Zhang S. paper collection is the Zhang S. paper representation, including paper 1 representation, paper 2 representation, paper 3 representation, and paper 4 representation. The specific SPECTER model calculation process is as follows:

[0049] E = M([T1,T2,…,T) i ]); (Formula 1)

[0050] Among them, T i This represents the text information of a certain field, M represents the SPECTER model, and E is the output feature vector.

[0051] However, at the same time, pre-trained models such as BERT or SPECTER cannot model discretized and structured information. For example, effective information such as the author's institution and co-author information needs to be modeled by building rule-based models to obtain feature vectors.

[0052] Furthermore, a distributed representation is obtained for patents in the research results dataset, that is, feature vectors are extracted from patents in a set of patents with the same name. For data such as patents without citation relationships, at least the following feature information is included: inventor information and patent information. Among them, inventor information includes the inventor's name, the inventor's institution, and information on co-inventors; patent information includes the patent title, patent summary, and patentee.

[0053] It should be noted that distributed representation of patents is more difficult than distributed representation of academic papers, meaning that feature vector extraction for patents is more complex. This is mainly because current pre-trained models for distributed representation of patents generally have poor training performance. Furthermore, due to the large time span of patents, textual information from different periods may exhibit issues such as missing fields and semantic ambiguity. Different countries and regions have different requirements and settings for patents. For example, the "patentee" field in early patents mostly contained personal names, but in recent years, this field has begun to include a large number of institutional names. In other words, while models exist for extracting patent feature vectors in some cases, these models often borrow from feature vector extraction models for other document types (such as academic papers). While they can extract vectors to some extent, they fail to consider the differences between patents and other document types. These differences are mainly reflected in the significant changes in patent structure and content over time, geographical variations, and language differences, making it impossible to fully and accurately extract the important features of the patent.

[0054] Due to the aforementioned issues, this embodiment employs unsupervised TF-IDF text representation for the distributed representation of patents, specifically using the TF-IDF model to extract patent feature vectors. TF-IDF is a commonly used text weighting technique, where TF stands for Term Frequency, representing the proportion of a particular word among all words in an article; and IDF stands for Inverse Document Frequency, representing the logarithm of the proportion of articles containing that word. The calculation process of the TF-IDF model includes the following:

[0055]

[0056]

[0057]

[0058] tfidf i,j =tf i,j *idf i ;(Formula 5)

[0059] Where j represents the article, i represents the words in the article, k refers to the total number of words in article j, and tf i,j |D| represents the frequency of each word in the article, |D| represents the total number of articles in the corpus, and |{j:t i ∈d j | represents the number of articles in the collection that contain this word, idf i Indicates the inverse document frequency of word i.

[0060] Unsupervised TF-IDF models are used to extract feature vectors from patents. For example, the distributed representation output by the Zhang S. patent set is the Zhang S. patent representation, which includes representations for patent 1, patent 2, and patent 3. Extracting feature vectors based on the TF-IDF model avoids obtaining erroneous and redundant information.

[0061] Step S14: Calculate the paper feature vector and patent feature vector based on the similarity algorithm and the rule-based model respectively to obtain the paper similarity matrix and the patent similarity matrix.

[0062] Specifically, a similarity algorithm is used to calculate the semantic similarity of the paper feature vector and the patent feature vector, respectively, to obtain a first similarity matrix of the paper feature vector and a first similarity matrix of the patent feature vector; a rule-based model is used to determine the similarity of the paper feature vector and the patent feature vector, respectively, to obtain a second similarity matrix of the paper feature vector and a second similarity matrix of the patent feature vector; a paper similarity matrix is ​​constructed based on the first similarity matrix of the paper feature vector and the second similarity matrix of the paper feature vector; a patent similarity matrix is ​​constructed based on the first similarity matrix of the patent feature vector and the second similarity matrix of the patent feature vector.

[0063] The similarity algorithm can utilize Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, cosine similarity, Mahalanobis distance, Pearson correlation coefficient, Jaccard similarity coefficient, etc. In this embodiment, the cosine similarity algorithm is preferably used to calculate the similarity matrix between the feature vectors of the paper and the feature vectors of the patent. Based on vectors x and y, the calculation formula for the cosine similarity algorithm is as follows:

[0064]

[0065] It should be noted that in the name disambiguation task, the use of cosine similarity algorithm as the similarity calculation method is based on the premise that the calculated feature vectors only involve differences between plain text content such as titles and abstracts. If feature vectors contain fields such as author affiliation information or co-author information, this embodiment uses a rule-based model. For example, the predefined rules of the rule-based model can be as follows:

[0066] Scenario 1: If research result x i With scientific research results x j If research results are published in the same year but at completely different institutions, their similarity is set to 0, indicating that research result x... i With scientific research results x j They inevitably belong to different people.

[0067] Scenario 2: If research result x i With scientific research results x j If the institutions affiliated with the authors of the research work overlap, and the co-authors of the research work also share commonalities, then it indicates that research work x... i With scientific research results x j It is possible that the same person from the same institution has a similarity score of 1, representing research achievement x. i With scientific research results x j from the same scholar.

[0068] Scenario 3: For research findings with citation relationships, such as name disambiguation tasks involving papers, the use of citation and cited relationships significantly impacts the disambiguation results. Specifically, two papers belonging to the same research field tend to have higher similarity, so a threshold can be set to determine whether they belong to the same field. Furthermore, articles in the same field are highly likely to cite other related papers. Based on this, if co-authors exist, it's almost certain that the two articles are from the same scholar, and their similarity is set to 1. Specific implementation examples of the above predefined rules are as follows:

[0069] Scenario 1:

[0070] Sim(x i ,x j ) = 0;

[0071] Scenario 2:

[0072] Sim(x i ,x j ) = 1;

[0073] Scenario 3:

[0074] and Cos(x i ,x j Shreshhold do;

[0075] Sim(x i ,x j ) = 1.

[0076] A model-based similarity matrix is ​​used to measure the degree of similarity between two or more objects. In this matrix, each row and column represents an object, and the elements represent the similarity or distance between corresponding objects. In the name disambiguation task, the similarity matrix is ​​an N*N matrix, where N is the number of papers in a set of papers with the same name or the number of patents in a set of patents with the same name. This matrix is ​​a pair-to-pair similarity matrix. Furthermore, from a time complexity perspective, the construction of the similarity matrix has a time complexity of O(N^2). 2 Regarding the representation of scientific research results, the similarity between all pairs of scientific research results must be calculated and judged.

[0077] Step S15: Process the paper similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the paper disambiguation result; and process the patent similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the patent disambiguation result.

[0078] It should be noted that after obtaining the point-to-point similarity matrix, only clustering operations need to be performed on the similarity matrix to complete the disambiguation task. Therefore, in this embodiment, unsupervised clustering algorithms are used for name disambiguation. The unsupervised clustering algorithms selected in this invention are nearest neighbor propagation algorithm and hierarchical clustering algorithm.

[0079] The nearest neighbor propagation algorithm treats all sample points as cluster centers, then reconstructs a similarity network from the lines connecting every two sample points, extracting a similarity matrix. Finally, it calculates the cluster centers of each class by evaluating the availability and responsibility of each edge in the network. A specific implementation example is as follows:

[0080] Input:

[0081] X = {x1, x2, ..., x} n}, the vector representation of the sample;

[0082] m, the number of given cluster categories;

[0083] λ, the attenuation coefficient, controls the convergence effect of the algorithm;

[0084] Output:

[0085] C = {c1, c2, c3, ..., c} m}, the m classes obtained after clustering;

[0086] Initialize Responsibility Matrix R=[r(i,j)];

[0087] Initialize Availability Matrix A=[a(i,j)]

[0088] Initialize Result C={c1,c2,c3,…,c n}

[0089] while R and A change do;

[0090]

[0091]

[0092] r t+1 (i,k)=λ*r t (i,k)+(1-λ)r t+1 (i,k);

[0093] a t+1 (i,k)=λ*a t (i,k)+(1-λ)a t+1 (i,k);

[0094] Choose argmax(r+a)as a cluster center.Update C;

[0095] return C.

[0096] Furthermore, hierarchical clustering algorithms are divided into two types: agglomerative hierarchical clustering and categorical hierarchical clustering. This embodiment preferably uses the hierarchical agglomerative clustering algorithm (HAC clustering algorithm). The idea behind the HAC clustering algorithm is: first, treat each sample as a separate class and calculate the distance between each pair of samples; then, merge the two classes with the smallest distance into a new class; then, recalculate the distance between the new class and all other classes; repeat the above operation until all classes are finally merged into one class. The advantage of using the HAC clustering algorithm is that it can display the clustering of sample points at different scales (hierarchies). The specific implementation process is as follows:

[0097]

[0098] For example, the final disambiguation results for papers include Zhang S.1 having papers 1 and 4, and Zhang S.2 having papers 2 and 3. The disambiguation results for patents include Zhang S.1 having patent 1, Zhang S.2 having patent 2, and Zhang S.3 having patent 3.

[0099] Step S16: Merge the paper dataset in the paper disambiguation results with the corresponding scholar's patent dataset in the patent disambiguation results to obtain the research results dataset of each scholar.

[0100] After name disambiguation is completed for all types of research output data, research output merging is performed based on the similarity between clusters. Once name disambiguation is complete, each cluster in research output format A is merged with the cluster in research output format B that has the highest similarity. For example, the merged result might be that Zhang S.1's research output includes Paper 1, Paper 4, and Patent 2; Zhang S.2's research output includes Paper 2, Paper 3, and Patent 3; and Zhang S.3's research output includes Patent 1. Specifically, this involves merging scholars' papers and patents. This step requires considering a combination of various rules, which can be set as follows:

[0101] Rule 1: If the similarity between a person's name and an organization's name is too low, they will not be merged.

[0102] Rule 2: If the similarity between the person's name and the organization's name exceeds the threshold, the degree of overlap between the collaborators and the collaborating organizations will be considered to determine whether to merge.

[0103] For example, taking a scholar named Li Fan as an example, Table 1 shows the results after performing name disambiguation and results merging algorithms on "LiF." of "Fudan University". The results show that there are two LiF.s within "Fudan University", one of whom has four papers and two patents, and the other has one paper and two patents.

[0104] Table 1. Results of disambiguation and merging of research findings for scholars named Li Fan

[0105]

[0106] It should be emphasized that the name disambiguation and merging method of research results proposed in this application is not only applicable to the field of scientific research, but can also be extended to other fields, such as medicine and law, providing an effective solution to the problem of identical names in various fields. Through accurate name disambiguation and merging, the citation and retrieval of literature are more precise and efficient, providing researchers with more convenient and accurate literature information. By merging the results of authors with the same name, a more complete and accurate scientific knowledge graph can be constructed, providing richer information and more accurate correlations for academic research and discipline development. Combining various information such as text features, collaboration relationships, institutional information, and citation relationships makes disambiguation and merging more comprehensive and integrated, improving the accuracy and robustness of the method. It can effectively identify the identity of researchers with the same name, avoiding different research results being incorrectly attributed to the same author, thereby ensuring the accuracy and credibility of research results.

[0107] To facilitate the demonstration of the name disambiguation and results merging methods of the research results of this application, the following specific embodiments are provided for illustration:

[0108] To demonstrate the effectiveness of this invention, six classic datasets from the personal name disambiguation task are introduced, including those from academic websites such as Aminer, ArnetMiner, INSPIRE, KISTI, QIAN, and SCAD-zbMATH. All six datasets contain training and testing sets, and some datasets also include validation sets for training. Dataset details are shown in Table 2.

[0109] Table 2 Information on the disambiguation benchmark dataset for names

[0110] Aminer 153147 600 31848 ArnetMiner 7067 103 1512 INSPIRE 265497 12458 14996 KISTI 36447 881 6856 QIAN 6542 580 1188 SCAD-zbMATH 12289 2136 2334

[0111] Table 3 shows that the name disambiguation system of the present invention significantly improves the disambiguation speed and achieves satisfactory disambiguation results without manual intervention.

[0112] Table 3 shows the performance of this invention on publicly available benchmark datasets.

[0113]

[0114] like Figure 3 The diagram illustrates the structure of a scientific research results disambiguation and results merging system 300 according to an embodiment of the present invention. The system 300 includes:

[0115] The scientific research results dataset acquisition module 301 is used to acquire the scientific research results dataset of the names to be disambiguated; the scientific research results dataset includes at least a paper dataset and a patent dataset;

[0116] Preprocessing module 302 is used to preprocess and divide the scientific research results dataset based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name.

[0117] The feature vector acquisition module 303 is used to extract the paper feature vectors of the papers in the same-name paper set based on the SPECTER model; and to extract the patent feature vectors of the patents in the same-name patent set based on the TF-IDF model.

[0118] The similarity matrix acquisition module 304 is used to calculate the paper feature vector and the patent feature vector based on the similarity algorithm and the rule-based model, respectively, to obtain the paper similarity matrix and the patent similarity matrix;

[0119] The disambiguation result acquisition module 305 is used to process the paper similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the paper disambiguation result; and to process the patent similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the patent disambiguation result;

[0120] The results merging module 306 is used to merge the paper dataset in the paper disambiguation results with the corresponding scholar's patent dataset in the patent disambiguation results to obtain the research results dataset of each scholar.

[0121] It should be understood that the division of the various modules or units in the above system is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules or units can be implemented entirely in software through processing element calls; they can be implemented entirely in hardware; or some modules or units can be implemented by processing element calls to software, while others are implemented in hardware.

[0122] For example, these modules or units can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to form a system-on-a-chip (SOC).

[0123] Since the implementation method of the scientific research results name disambiguation and results merging system in this embodiment is similar to the implementation method of the scientific research results name disambiguation and results merging method provided above, it will not be described again.

[0124] like Figure 4 The diagram illustrates the structure of an electronic terminal in one embodiment of this application. The electronic terminal 40 provided in this example includes a memory 41 and a processor 42. The memory 41 stores a computer program; the processor 42 runs the computer program to implement the aforementioned method for disambiguating names and merging research results.

[0125] Optionally, the number of memories 41 can be one or more, and the number of processors 42 can be one or more.

[0126] Optionally, the processor 42 in the electronic terminal 40 will perform the following... Figure 1 The steps described involve loading one or more instructions corresponding to the process of an application into memory 41, and having the processor 42 run the application stored in the first memory 41, thereby realizing various functions in the method for disambiguating names and merging results of scientific research achievements.

[0127] Optionally, the memory 41 may include, but is not limited to, high-speed random access memory and non-volatile memory. For example, one or more disk storage devices, flash memory devices, or other non-volatile solid-state storage devices; the processor 42 may include, but is not limited to, a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0128] Optionally, the processor 42 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0129] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a method for disambiguating names and merging results in scientific research.

[0130] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented using computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0131] In the embodiments provided in this application, the computer-readable and writable storage medium may include read-only memory, random access memory, EEPROM, CD-ROM or other optical disc storage devices, disk storage devices or other magnetic storage devices, flash memory, USB flash drive, portable hard drive, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a computer. Additionally, any connection may be appropriately referred to as a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of the medium. However, it should be understood that computer-readable and writable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but are intended for non-transient, tangible storage media. The disks and optical discs used in the application include compact discs (CDs), laser discs, optical discs, digital multifunction discs (DVDs), floppy disks, and Blu-ray discs, where disks typically copy data magnetically, while optical discs use lasers to copy data optically.

[0132] In summary, this application provides a method, system, terminal, and medium for disambiguation and merging of research results. The method is applicable not only to scientific research but also to other fields such as medicine and law, providing an effective solution to the problem of identical names in various fields. Through accurate disambiguation and merging, literature citation and retrieval become more precise and efficient, providing researchers with more convenient and accurate literature information. By merging results from authors with the same name, a more complete and accurate scientific knowledge graph can be constructed, providing richer information and more accurate correlations for academic research and disciplinary development. By combining textual features, collaboration relationships, institutional information, and citation relationships, the disambiguation and merging become more comprehensive and integrated, improving the accuracy and robustness of the method. It effectively identifies the identities of authors with the same name, preventing research results from being incorrectly attributed to different authors, thus ensuring the accuracy and credibility of research results. Therefore, this application effectively overcomes the various shortcomings of existing technologies and has high industrial application value.

[0133] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for disambiguating names and merging research results, characterized in that, include: Obtain the dataset of research findings containing the names to be disambiguated; The research results dataset includes at least a paper dataset and a patent dataset; The research results dataset was preprocessed and partitioned based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name. The paper feature vectors of the papers in the same-name paper set are extracted based on the SPECTER model; the patent feature vectors of the patents in the same-name patent set are extracted based on the TF-IDF model. The paper feature vector and patent feature vector are calculated based on similarity algorithm and rule-based model respectively to obtain paper similarity matrix and patent similarity matrix; The paper similarity matrix is ​​processed using the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain paper disambiguation results; and the patent similarity matrix is ​​processed using the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain patent disambiguation results; The paper dataset from the paper disambiguation results is merged with the corresponding scholar's patent dataset from the patent disambiguation results to obtain the research results dataset of each scholar.

2. The method for disambiguating names and merging research results according to claim 1, characterized in that, The preprocessing and partitioning of the scientific research dataset based on the LN-FI criterion to obtain a set of papers and patents with the same name specifically includes: The names of scholars in the research results dataset are standardized based on the LN-FI criterion to obtain a unified format for the names of scholars. The research results dataset is divided into homonyms based on the unified scholar name format to obtain the sets of papers and patents with the same name for each scholar with the same name in the research results dataset.

3. The method for disambiguating names and merging research results according to claim 1, characterized in that, The calculation formula for the SPECTER model is as follows: E=M([T1,T2,…,T i ]); Among them, T i This represents the text information of a certain field, M represents the SPECTER model, and E is the output feature vector.

4. The method for disambiguating names and merging research results according to claim 1, characterized in that, The calculation process of the TF-IDF model includes the following: tfidf i,j =tf i,j *idf i ; Where j represents the article, i represents the words in the article, k refers to the total number of words in article j, and tf i,j |D| represents the frequency of each word in the article, |D| represents the total number of articles in the corpus, and |{j:t i ∈d j | represents the number of articles in the collection that contain this word, idf i Indicates the inverse document frequency of word i.

5. The method for disambiguating names and merging research results according to claim 1, characterized in that, The similarity-based algorithm and rule-based model calculate the feature vectors of the paper and the patent, respectively, to obtain the paper similarity matrix and the patent similarity matrix, specifically including: The semantic similarity of the paper feature vector and the patent feature vector is calculated using a similarity algorithm to obtain the first similarity matrix of the paper feature vector and the first similarity matrix of the patent feature vector. A rule-based model is used to determine the similarity between the feature vectors of the paper and the feature vectors of the patent, respectively, to obtain the second similarity matrix of the feature vectors of the paper and the second similarity matrix of the feature vectors of the patent. A paper similarity matrix is ​​constructed based on the first similarity matrix and the second similarity matrix of the paper's feature vectors. A patent similarity matrix is ​​constructed based on the first similarity matrix and the second similarity matrix of the patent feature vectors.

6. The method for disambiguating names and merging research results according to claim 1, characterized in that, The calculation process of the HAC clustering algorithm includes the following: Treat each data point in the sample set as a cluster; Calculate the distance between any two clusters, and merge the two closest or most similar clusters into a new cluster; Calculate the distance between the new cluster and all other clusters; Repeat the above steps until the iteration termination condition is met.

7. A system for disambiguating names and merging research results, characterized in that, include: The research results dataset acquisition module is used to acquire the research results dataset containing the names to be disambiguated. The research results dataset includes at least a paper dataset and a patent dataset; The preprocessing module is used to preprocess and divide the research results dataset based on the LN-FI criterion to obtain a set of papers with the same name and a set of patents with the same name. The feature vector acquisition module is used to extract the paper feature vectors of the papers in the same-name paper set based on the SPECTER model; and to extract the patent feature vectors of the patents in the same-name patent set based on the TF-IDF model. The similarity matrix acquisition module is used to calculate the paper feature vector and the patent feature vector based on the similarity algorithm and the rule-based model, respectively, to obtain the paper similarity matrix and the patent similarity matrix; The disambiguation result acquisition module is used to process the paper similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the paper disambiguation result; and to process the patent similarity matrix based on the nearest neighbor propagation algorithm and the HAC clustering algorithm to obtain the patent disambiguation result; The results merging module is used to merge the paper dataset from the paper disambiguation results with the corresponding scholar's patent dataset from the patent disambiguation results to obtain the research results dataset of each scholar.

8. The system for disambiguating names and merging research results according to claim 7, characterized in that, The preprocessing module includes: The standardization processing module is used to standardize the names of scholars in the research results dataset based on the LN-FI criterion to obtain a unified format for the names of scholars. The same-name segmentation module is used to segment the scientific research results dataset according to the unified scholar name format, so as to obtain the same-name paper set and same-name patent set of each scholar in the scientific research results dataset.

9. An electronic terminal, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is configured to execute the computer program stored in the memory, so that the electronic terminal performs the method for disambiguating names and merging research results as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for disambiguating names and merging research results as described in any one of claims 1 to 6.

Images