Abnormal reference detection method, device and equipment and computer storage medium

Abstract

Embodiments of the present application provide a method, device and equipment for detecting abnormal citation, and a computer storage medium, and relate to the field of text detection. The method comprises: obtaining the number of citations between each pair of journals in a target field; for each pair of journals, determining a strength set of the pair of journals according to the number of citations between the pair of journals and the total number of citations between all journals; clustering the journals in the target field according to the strength set of each pair of journals to obtain a first set; for each pair of papers in the first set that have a citation relationship, obtaining the semantic similarity between the pair of papers and the first number of citations of the cited paper by papers in all journals in the first set, and determining the abnormal probability of the citation relationship of the pair of papers according to the semantic similarity and the first number of citations; and for each first set, determining whether the first set has abnormal citation according to the abnormal probability of the pair of papers in the first set. The present application improves the accuracy of identifying abnormal citation based on non-academic motivation.

Description

Technical Field

[0001] This application relates to the field of text detection technology, and more specifically, to a method, apparatus, device, and computer storage medium for detecting abnormal references. Background Technology

[0002] In academic evaluation practice, citation metrics such as journal impact factor are widely used to measure a journal's academic influence. Some journals manipulate evaluation metrics such as impact factor by forming "citation cartels" with other journals to artificially increase each other's citation counts. This collective citation manipulation typically manifests as: a high-density network of mutual citations among multiple journals, with internal citation frequencies significantly higher than normal academic collaboration levels, and the cited content often being unrelated or only marginally relevant to the paper's topic.

[0003] However, there are anomalous citations in the relevant technologies that cannot accurately identify non-academic motives. Summary of the Invention

[0004] This application provides a method, apparatus, device, and computer storage medium for detecting aberrant citations, which addresses the technical problem of being unable to accurately identify aberrant citations with non-academic motives.

[0005] According to a first aspect of the embodiments of this application, a method for detecting abnormal citations is provided. The method includes: obtaining the number of citations between each journal pair in a target field; for each journal pair, determining a strength set of the journal pair based on the number of citations between the journal pairs and the sum of the number of citations between all journals; the strength set is used to characterize the strength of mutual citations between papers published in the two journals included in the journal pair, and the strength set includes the strength of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal. Based on the intensity set of each journal pair, the journals in the target domain are clustered to obtain at least one first set; each first set includes at least one journal. For each first set, for each paper pair in the first set that has a citation relationship, obtain the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by all journals in the first set. Determine the abnormal probability of the citation relationship of the paper pair based on the semantic similarity and the first number of citations. For each first set, determine whether there are aberrant citations in the first set based on the aberration probability of all paper pairs with citation relationships within the first set.

[0006] In one possible implementation, the strength of the citing journal's citation of the cited journal is determined by the following method: The system obtains the second total number of times the citing journal cites the cited journal, the first total number of times the citing journal cites all journals in the target field, the second total number of times the cited journal is cited by each journal in the target field, and the third total number of times each journal in the target field cites the cited journal. The expected citation count is determined based on the first total number of citations, the second total number of citations, and the third total number of citations. The expected citation count refers to the number of times the citing journal cites the cited journal in the third total number of citations. The intensity of the citing journal's citation of the cited journal is determined based on the expected number, the second number, and the first total number.

[0007] In another possible implementation, the first probability of the citing journal citing all journals in the target field in the third total number of citations is determined based on the ratio of the first total number of citations to the third total number of citations. Based on the ratio of the second total number of citations to the third total number of citations, determine the second probability that the cited journal will be cited by other journals in the target field in the third total number of citations; Based on the first and second probabilities, determine the third probability of the citing journal citing the cited journal in the third total number of citations; The expected number of citations is determined based on the third probability and the third total number of citations.

[0008] In another possible implementation, the fourth probability of the citing journal citing the cited journal in the first total number of citations is determined based on the ratio between the second number and the first total number of citations. The ratio between the second citation count and the expected citation count determines the degree of difference between the actual number of times the citing journal cites the cited journal and the expected number of citations. The strength of the citing journal's citation of the cited journal is determined based on the degree of difference and the fourth probability.

[0009] In another possible implementation, for each journal pair, if both intensities in the intensity set of the journal pair are greater than a preset first intensity threshold, and the sum of the two intensities is greater than a preset second intensity threshold, then the journal pair is determined to have a weight, and the first mean of the two intensities is used as the weight of the journal pair. Construct a topology graph of the target domain based on all journals and weighted journal pairs within the target domain. The topology graph includes nodes and edges. A node represents a journal, and the two nodes in a weighted journal pair are connected by an edge. The weight of the journal pair is the weight of the corresponding edge. Cluster all journals in the target domain based on the topology graph to obtain at least one first set.

[0010] In another possible implementation, after obtaining the first set, the first total number of all journals in the first set is obtained, and the total number of journal pairs in the first set is determined based on the first total number. Obtain the second total number of journal pairs in the first set whose two intensities are both greater than a preset first intensity threshold, and whose sum of the two intensities is greater than a preset second intensity threshold; The ratio of the second total number to the total logarithm is used as the degree of closeness among journals within the first set; If it is determined that the first total quantity is less than a preset quantity threshold and the density is greater than a preset density threshold, then at least one of the first sets in the first set is retained.

[0011] In yet another possible implementation, the first abstract text of the citing paper and the second abstract text of the cited paper are extracted; Obtain the cosine similarity between the first and second summary texts; Extract at least one first keyword from the citing paper and at least one second keyword from the cited paper; Get the number of identical keywords in at least one primary keyword and at least one secondary keyword; get the total number of keyword types in at least one primary keyword and at least one secondary keyword; get the percentage of identical keywords in the total number of types. The semantic similarity is obtained by weighting and summing the cosine similarity and the ratio according to the preset weight coefficients.

[0012] In another possible implementation, the first anomaly probability is determined based on semantic similarity; the first anomaly probability is linearly negatively correlated with semantic similarity. The adjustment coefficient is determined based on the first number; the adjustment coefficient has a monotonically increasing relationship with the first number, and the increment gradually decreases; By adjusting the coefficients to adjust the first anomaly probability, the anomaly probability of the paper's citation relationship anomaly is obtained.

[0013] In yet another possible implementation, the second mean of the outlier probabilities of all paper pairs with citation relationships within the first set is obtained; For each paper pair with a citation relationship, if the probability of an anomaly in the paper pair is greater than a preset anomaly probability threshold, then the paper pair is considered an anomaly paper pair. Get the third total number of all abnormal paper pairs in the first set and the fourth total number of all paper pairs in the first set; If the second mean is greater than the preset mean threshold, and the ratio of the third total quantity to the fourth total quantity is greater than the preset ratio threshold, then it is determined that there is an abnormal reference in the first set.

[0014] In yet another possible implementation, after determining that the first set contains an abnormal reference, output at least one of the following: The names of the journals in the first set; At least one paper pair with the highest anomalous probability in the first set; The topological graph corresponding to the first set; The topology graph includes nodes and edges. A node represents a journal in the first set. The nodes representing the two journals in a pair of journals with anomalous paper pairs in the first set are connected by edges.

[0015] According to a second aspect of the embodiments of this application, an abnormal reference detection device is provided, the device comprising: The acquisition module is used to acquire the citation count between each journal pair in the target field. For each journal pair, the intensity set of the journal pair is determined based on the citation count between the journal pairs and the total citation count between all journals. The intensity set is used to characterize the intensity of mutual citation between the papers published in the two journals included in the journal pair. The intensity set includes the intensity of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal. The clustering module is used to cluster journals in the target domain based on the intensity set of each journal pair to obtain at least one first set; each first set includes at least one journal. The determination module is used to, for each first set, for each paper pair in the first set that has a citation relationship, obtain the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by papers in all journals in the first set, and determine the abnormal probability of the citation relationship of the paper pair based on the semantic similarity and the first number of citations. The judgment module is used to determine whether there are abnormal citations in each first set based on the abnormal probability of all paper pairs with citation relationships in the first set.

[0016] According to a third aspect of the embodiments of this application, an electronic device is provided, the electronic device including a memory, a processor and a computer program stored in the memory, wherein the processor executes the program to implement the steps of the method provided in the first aspect.

[0017] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps of the method provided in the first aspect.

[0018] According to a fifth aspect of the present application, a computer program product is provided, the computer program product including computer instructions stored in a computer-readable storage medium, wherein when a processor of a computer device reads the computer instructions from the computer-readable storage medium, the processor executes the computer instructions, causing the computer device to perform steps implementing the method provided in the first aspect.

[0019] The beneficial effects of the technical solutions provided in this application are: By obtaining the citation count between each journal pair within the target field, and for each journal pair, determining the intensity set of the journal pair based on the citation count between the journal pairs and the total citation count between all journal pairs, the intensity of mutual citations between the papers published in the two journals included in each journal pair can be quantified by using the citation count between the two journals in the journal pair and the total citation count between all journal pairs.

[0020] By clustering journals in the target field according to the intensity set of each journal pair, at least one first set including at least one journal is obtained. This realizes the ability to group journals with high mutual citation intensity into a first set based on the mutual citation intensity between journal pairs, so as to perform abnormal citation detection on each first set in the future. For each first set, for each paper pair with a citation relationship in the first set, the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by all journals in the first set are obtained. The abnormal probability of the citation relationship of the paper pair is determined based on the semantic similarity and the first number of citations. The content relevance between the citing paper and the cited paper is quantified by the semantic similarity. Combined with the content relevance and the first number of times the citing paper is cited by all journals, the abnormal probability of the citation relationship of the paper pair is determined. This enables more accurate identification of citation behavior in paper pairs that is not motivated by academic research. For each set, the presence of aberrant citations in the first set is determined by the aberration probability of all paper pairs with citation relationships within the first set. By combining the aberration probabilities of all paper pairs with citation relationships within the first set, the presence of aberrant citations in the first set as a whole is determined. Thus, the determination of whether the first set is a citation manipulation alliance with citation manipulation behavior can be made based on the judgment result. This solves the problem of distinguishing between reasonable academic cooperation (such as a group of journals with similar disciplines) and maliciously formed citation manipulation alliances. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0022] Figure 1 A flowchart illustrating an abnormal reference detection method provided in an embodiment of this application; Figure 2 A flowchart illustrating the method for determining the strength of citations from a citing journal to a cited journal in an abnormal citation detection method provided in this application embodiment; Figure 3 A schematic diagram of a method for determining the expected number of references in an abnormal reference detection method provided in an embodiment of this application; Figure 4 A flowchart illustrating the intensity determination method in an abnormal reference detection method provided in an embodiment of this application; Figure 5 A flowchart illustrating the method for obtaining the first set in an abnormal reference detection method provided in an embodiment of this application; Figure 6 A flowchart illustrating the filtering method for the first set in an abnormal reference detection method provided in an embodiment of this application; Figure 7 A schematic diagram of the structure of an abnormal reference detection device provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0024] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.”

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0026] The relevant technologies are explained below: Methods for detecting anomalous references in related technologies fall into the following two categories: 1. Based on traditional network topology analysis methods. By constructing the citation network of papers or journals, and using methods such as graph theory, community detection, or random block models to analyze the topological characteristics of the citation network, dense subgraphs with abnormal citation structures can be discovered to identify citation manipulation alliances.

[0027] 2. Graph representation learning-based methods: Deep learning techniques are used to map nodes (or edges) in the citation network of papers or journals to a low-dimensional vector space. Deep reinforcement learning, graph neural networks, graph embedding, network perturbation and other techniques are used to automatically learn complex patterns in order to deal with the problem of identifying citation manipulation alliances in large-scale citation networks.

[0028] However, the relevant methods have the following drawbacks: 1. It is difficult to distinguish between legitimate academic collaborations (such as a group of journals with similar disciplines) and malicious citation manipulation alliances; 2. The identification results are only probabilistic anomalies, lacking interpretable evidence, and cannot output specific paper pairs with citation anomalies; 3. The semantic relevance of the cited content was not considered, and the typical manipulation feature of "frequent cross-citation despite unrelated content" could not be identified.

[0029] In view of at least one of the above-mentioned technical problems or areas for improvement in the related technologies, this application proposes a method for detecting abnormal citations. This method obtains the number of citations between each journal pair in the target field. For each journal pair, the intensity set of the journal pair is determined based on the number of citations between the journal pairs and the total number of citations between all journal pairs. This method quantifies the intensity of mutual citations between papers published in the two journals included in each journal pair by using the number of citations between the two journals in the journal pair and the total number of citations between all journal pairs.

[0030] By clustering journals in the target field according to the intensity set of each journal pair, at least one first set including at least one journal is obtained. This realizes the ability to group journals with high mutual citation intensity into a first set based on the mutual citation intensity between journal pairs, so as to perform abnormal citation detection on each first set in the future. For each first set, for each paper pair with a citation relationship in the first set, the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by all journals in the first set are obtained. The abnormal probability of the citation relationship of the paper pair is determined based on the semantic similarity and the first number of citations. The content relevance between the citing paper and the cited paper is quantified by the semantic similarity. Combined with the content relevance and the first number of times the citing paper is cited by all journals, the abnormal probability of the citation relationship of the paper pair is determined. This enables more accurate identification of citation behavior in paper pairs that is not motivated by academic research. For each set, the presence of aberrant citations in the first set is determined by the aberration probability of all paper pairs with citation relationships within the first set. By combining the aberration probabilities of all paper pairs with citation relationships within the first set, the presence of aberrant citations in the first set as a whole is determined. Thus, the determination of whether the first set is a citation manipulation alliance with citation manipulation behavior can be made based on the judgment result. This solves the problem of distinguishing between reasonable academic cooperation (such as a group of journals with similar disciplines) and maliciously formed citation manipulation alliances.

[0031] The technical solutions of this application and their effects are described below through several exemplary embodiments. It should be noted that the following embodiments can be referenced, borrowed from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0032] This application provides a method for detecting abnormal references, such as... Figure 1 As shown, the method includes: S101, obtain the citation count between each journal pair in the target field. For each journal pair, determine the intensity set of the journal pair based on the citation count between journal pairs and the sum of the citation counts between all journals.

[0033] In the embodiments of this application, the target field refers to any subject area, and the journal pair consists of two journals within the target field. For example, if the journal pair includes journal A and journal B, then the journal pair includes the following two cases: when the citing journal is A, the cited journal is B, and when the citing journal is B, the cited journal is A.

[0034] In the embodiments of this application, the number of citations between journal pairs refers to the number of times that papers published in the two journals in the journal pair cite each other. For example, when a journal pair consists of a first journal and a second journal, the number of citations between journal pairs includes: the number of times the first journal, as the citing journal, cites the second journal, as the cited journal, and the number of times the second journal, as the citing journal, cites the first journal, as the cited journal.

[0035] In one example, the journal pair includes journal A and journal B. Journal A publishes papers A and B, and journal B publishes papers D, E, and F. Paper A cites papers D and E, which is recorded as journal A citing journal B twice. Paper F cites paper B, which is recorded as journal B citing journal A once. Therefore, the number of citations between the journal pairs is: journal A cites journal B twice, and journal B cites journal A once.

[0036] In this embodiment of the application, the number of citations between journal pairs is obtained by statistically analyzing the citation relationships between the papers published in the two journal pairs. For example, the document metadata within a preset time period is obtained from an academic paper and citation database. The document metadata includes the citation relationships between papers. The citation relationships between the papers included in the two journal pairs are statistically analyzed to obtain the number of citations between the journal pairs.

[0037] In the embodiments of this application, the intensity set is used to characterize the strength of mutual citation between papers published in the two journals included in the journal pair. The intensity set includes the strength of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal.

[0038] In this embodiment of the application, the total number of citations among all journals includes: the total number of citations among papers published in different journals and the total number of citations among different papers published in the same journal.

[0039] In the embodiments of this application, the greater the proportion of the number of times the citing journal cites the cited journal to the total number of citations among all journals, the greater the intensity of the citation of the cited journal by the citing journal. Therefore, for each journal pair, the intensity of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal can be obtained based on the number of citations between journal pairs and the total number of citations between all journal pairs.

[0040] S102, based on the intensity set of each journal pair, cluster the journals in the target domain to obtain at least one first set.

[0041] In the embodiments of this application, since the intensity set of each journal pair can characterize the intensity of mutual citation between the two journals included in the journal pair, the higher the intensity set, the higher the citation frequency between the two journals in the journal pair, and the higher the cohesion. When each intensity in the intensity set of the journal pair is higher than a certain intensity, it indicates that there may be abnormal citation between the two journals in the journal pair.

[0042] In this embodiment of the application, each first set includes at least one journal. Since the intensity of mutual citation between journals is too high, there may be abnormal citations between journals. Therefore, the first set is a candidate citation alliance. Journals in the target field are clustered according to the intensity set of each journal pair, that is, journals with high mutual citation intensity are grouped into a first set.

[0043] S103, For each first set, for each paper pair in the first set that has a citation relationship, obtain the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by all journals in the first set. Determine the abnormal probability of the citation relationship of the paper pair based on the semantic similarity and the first number of citations.

[0044] In this embodiment of the application, a paper pair refers to any two papers among all the papers published in all journals that make up the first set, and a paper pair with a citation relationship refers to a paper pair in which two papers cite each other.

[0045] In this embodiment of the application, the semantic similarity between the citing paper and the cited paper in the paper pair is obtained to determine the content relevance between the citing paper and the cited paper. The greater the semantic similarity, the higher the content relevance between the citing paper and the cited paper. The higher the content relevance, the greater the probability that the citation in the current paper pair is based on academic purposes.

[0046] In this embodiment of the application, the citation influence of the cited paper in the first set is determined by obtaining the first number of times the cited paper is cited by all journals in the first set. The more times it is cited, the greater the citation influence.

[0047] In this embodiment, since a lower semantic similarity indicates a lower content relevance between the citing and cited papers, the probability that the citing paper's purpose in citing the cited paper is not for academic purposes is higher. The more times a citing paper is cited by all papers in the first set, the greater its influence within the first set. When the content relevance between the citing and cited papers is lower, but the influence of the cited paper within the first set is greater, it further indicates a higher probability that the current paper has an anomalous citation. Therefore, the probability of an anomalous citation relationship between the paper and the cited paper is determined based on semantic similarity and the number of first citations.

[0048] S104. For each first set, determine whether there are abnormal citations in the first set based on the abnormal probability of all paper pairs with citation relationships within the first set.

[0049] In this embodiment of the application, the overall probability of abnormal citations in the first set is determined by statistically analyzing the abnormal probability of all paper pairs with citation relationships within the first set. When the overall probability is large, it can be said that the first set is a citation alliance with abnormal citations.

[0050] In this embodiment of the application, if the number of paper pairs with citation relationships is greater than a preset sampling threshold (ranging from 100 to 1000), then a sampling threshold number of paper pairs are randomly selected from all eligible paper pairs to determine the probability of anomalies.

[0051] In this embodiment of the application, the existence of abnormal citations in the first set can be determined based on the average of the abnormal probabilities of all paper pairs with citation relationships. When the average abnormal probability exceeds a preset probability threshold, the first set is considered to have abnormal citations.

[0052] In this embodiment of the application, it is also possible to determine whether the first set is a citation alliance with citation anomalies based on the distribution range of the abnormal probability of all papers with citation relationships in the first set. For example, the abnormal probability is divided into three ranges: 0-33%, 34%-67%, and 68%-100%. When the abnormal probability of more than half of the paper pairs is in the 68%-100% range, the first set is considered to have abnormal citations.

[0053] The abnormal citation detection method provided in this application obtains the citation count between each journal pair in the target field. For each journal pair, the intensity set of the journal pair is determined based on the citation count between the journal pairs and the total citation count between all journal pairs. This method quantifies the intensity of mutual citations between papers published in the two journals included in each journal pair by using the citation count between the two journals in the journal pair and the total citation count between all journal pairs.

[0054] By clustering journals in the target field according to the intensity set of each journal pair, at least one first set including at least one journal is obtained. This realizes the ability to group journals with high mutual citation intensity into a first set based on the mutual citation intensity between journal pairs, so as to perform abnormal citation detection on each first set in the future. For each first set, for each paper pair with a citation relationship in the first set, the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by all journals in the first set are obtained. The abnormal probability of the citation relationship of the paper pair is determined based on the semantic similarity and the first number of citations. The content relevance between the citing paper and the cited paper is quantified by the semantic similarity. Combined with the content relevance and the first number of times the citing paper is cited by all journals, the abnormal probability of the citation relationship of the paper pair is determined. This enables more accurate identification of citation behavior in paper pairs that is not motivated by academic research. For each set, the presence of aberrant citations in the first set is determined by the aberration probability of all paper pairs with citation relationships within the first set. By combining the aberration probabilities of all paper pairs with citation relationships within the first set, the presence of aberrant citations in the first set as a whole is determined. Thus, the determination of whether the first set is a citation manipulation alliance with citation manipulation behavior can be made based on the judgment result. This solves the problem of distinguishing between reasonable academic cooperation (such as a group of journals with similar disciplines) and maliciously formed citation manipulation alliances.

[0055] Based on the above embodiments, as an optional embodiment, another method for determining the strength of a citing journal's citation of a cited journal is provided, such as... Figure 2 As shown, the specific content is as follows: S201, obtain the second number of times the citing journal cites the cited journal, the first total number of times the citing journal cites all journals in the target field, the second total number of times the cited journal is cited by each journal in the target field, and the third total number of times each journal in the target field cites the other two journals. S202, determine the expected number of citations based on the first total number of citations, the second total number of citations, and the third total number of citations; the expected number of citations refers to the expected number of times the citing journal cites the cited journal in the third total number of citations. S203 determines the intensity of the citation of the cited journal by the citing journal based on the expected number, the second number, and the first total number.

[0056] In S201 of this application embodiment, the second number of times the citing journal cites the cited journal is obtained by counting the total number of times all papers published in the citing journal cite each paper published in the cited journal. The first total number of times the citing journal cites all journals in the target field is obtained by counting the total number of times all papers published in the citing journal cite each paper published in the target field. The second total number of times the cited journal is cited by each journal in the target field is obtained by counting the total number of times the papers published in the cited journal are cited by any paper in each journal in the target field. The third total number of times the journals in the target field cite each other is obtained by counting the total number of times the papers in any journal in the target field cite each other. It should be noted that the above-mentioned "two journals" includes the number of times the same journals cite each other.

[0057] In S202 of this application embodiment, when the third total number of citations among all journals is determined, the first total number of citations can be determined to determine the proportion of citations by the citing journal, and the second total number of citations can be determined to determine the proportion of citations by the cited journal. When the above two proportions are known, the expected number of times the citing journal cites the cited journal in the third total number of citations can be obtained, that is, the number of times the citing journal cites the cited journal under normal citation conditions.

[0058] In S203 of this application embodiment, under the known third total number of citations, the expected number of times the citing journal cites the cited journal and the actual second number of times the citing journal cites the cited journal can determine the degree of difference between the actual number of citations and the number of citations under normal citation conditions. Based on the second number and the first total number of citations, the proportion of the citing journal citing the cited journal in the total number of citations can also be determined. When the degree of difference is greater and the proportion is greater, the citation intensity of the corresponding journal pair is also stronger.

[0059] In the above scheme, the citation intensity between the citing and cited journals is determined by statistically analyzing the second number of citations by the citing journal, the first total number of citations by the citing journal to all journals in the target field, the second total number of citations by the cited journal to all journals in the target field, and the third total number of citations between any two journals in the target field. This achieves the quantification of the citation intensity between the citing and cited journals based on the third total number of citations in the entire target field, improving the accuracy of anomalous citation detection and providing reliable data support for subsequent anomalous citation detection.

[0060] Based on the above embodiments, as an optional embodiment, a method for determining the expected number of citations is provided, such as... Figure 3 As shown, the specific content is as follows: S301, Based on the ratio of the first total number of citations to the third total number of citations, determine the first probability of the citing journal citing all journals in the target field in the third total number of citations; S302, Based on the ratio of the second total number of citations to the third total number of citations, determine the second probability that the cited journal will be cited by other journals in the target field in the third total number of citations; S303, based on the first probability and the second probability, determine the third probability of the citing journal citing the cited journal in the third total number of citations; S304, determine the expected number of citations based on the third probability and the third total number of citations.

[0061] In S301 of this application embodiment, obtaining the ratio of the first total number of citations to the third total number of citations is equivalent to determining the proportion of citations by the citing journal among all journals in the third total number of citations among all journals, thereby determining the first probability of citation by the citing journal in the third total number of citations. In S302 of this application embodiment, obtaining the ratio of the second total number of citations to the third total number of citations is equivalent to determining the proportion of the cited journal being cited by all journals in the third total number of citations among all journals, thereby determining the second probability of the cited journal being cited in the third total number of citations. In S303 of this application embodiment, given the first probability of the citing journal performing a citation and the second probability of the cited journal being cited, the third probability of the citing journal citing the cited journal under normal citation conditions can be determined based on the product of the first and second probabilities. In S304 of this application embodiment, given the third probability of the citing journal citing the cited journal and the third total number of citations among all journals, the expected number of citations of the citing journal by the citing journal under normal citation conditions can be determined based on the product of the third probability and the third total number of citations.

[0062] In one example, the expected number of citations of the citing journal (i) to the citing journal (j) is obtained using the following formula. ;

[0063] Where i ≠ j, This represents the first total number of citations of journal i across all journals in the target field. Let be the second total number of times journal j is cited by all journals in the target field, and n be the total number of journals in the target field. This refers to the third total number of citations between any two journals within the target field.

[0064] In the above scheme, by obtaining the ratio of the first total number of citations to the third total number of citations in the citing journal, the first probability of citation by the citing journal is determined based on the actual citation situation. By obtaining the ratio of the second total number of citations to the third total number of citations in the cited journal, the second probability of citation by the cited journal is determined based on the actual citation situation. Based on the knowledge of the above two probabilities, the third probability of the citing journal citing the cited journal can be determined by multiplying the first and second probabilities. That is, the probability of the citing journal citing the cited journal under normal citation conditions. Given the third total number of citations and the third probability between any two journals in the target field, the expected number of citations by the citing journal under normal citation conditions within the third total number of citations in the target field can be determined. By calculating the expected number of citations based on the actual number of citations, a reference standard can be provided for subsequent calculations of the citation intensity between journal pairs, improving the accuracy of abnormal citations.

[0065] Based on the above embodiments, as an optional embodiment, a method for determining intensity is provided, such as... Figure 4 As shown, the specific content is as follows: S401, based on the ratio between the second number and the first total number, determine the fourth probability of the citing journal citing the cited journal in the first total number of citations; S402, determine the degree of difference between the number of times the citing journal actually cites the cited journal and the expected number of citations based on the ratio between the second number and the expected number of citations; S403 determines the strength of a citing journal's citation of a cited journal based on the degree of difference and the fourth probability.

[0066] In S401 of this application embodiment, the proportion of citations of the cited journal in all citations of the citing journal can be determined based on the ratio between the second number and the first total number of citations, thereby obtaining the fourth probability that the citing journal cites the cited journal in the first total number of citations.

[0067] In S402 of this application embodiment, since the second count is the number of times the citing journal actually cites the cited journal, and the expected citation count is the number of citations under normal citation conditions, by obtaining the ratio between the second count and the expected citation count, the degree of difference between the actual citation count and the expected citation count of the citing journal and the cited journal can be determined. The higher the degree of difference, the greater the intensity of the citation of the cited journal by the citing journal.

[0068] In S403 of this application embodiment, since the degree of difference reflects the difference between the actual citation and the expected citation, and the fourth probability reflects the probability that the citing journal cites the cited journal, the greater the degree of difference and the greater the citation of the cited journal by the citing journal, the stronger the citation strength of the corresponding journal pair.

[0069] In one example, the strength of the citing journal (i) citing the cited journal (j) is obtained using the following formula. :

[0070] Where i ≠ j, This represents the first total number of citations of journal i across all journals in the target field. The expected number of citations by the citing journal (i) to the cited journal (j), This indicates the second time journal i cites journal j, ε is a smoothing parameter with a value between 0.001 and 0.1 to prevent the logarithm from being infinite, and the function log is the natural logarithm.

[0071] In the above scheme, by obtaining the second number and the expected number of citations, the degree of difference between the actual number of citations and the expected number of citations under normal citation conditions is determined. The greater the difference, the higher the citation intensity of the journal pair. Based on the ratio between the second number and the first total number of citations, the proportion of the citing journal citing the cited journal to the total number of journals cited by the citing journal is determined. The greater the proportion, the greater the citation intensity between the journal pairs. Therefore, based on the degree of difference and the fourth probability, the intensity of the citation of the cited journal by the citing journal is determined, which realizes the reasonable quantification of citation intensity based on the actual citation situation and improves the accuracy of anomaly detection.

[0072] Based on the above embodiments, as an optional embodiment, a method for obtaining the first set is provided, such as... Figure 5 As shown, the specific content is as follows: S501, For each journal pair, if both intensities in the intensity set of the journal pair are greater than a preset first intensity threshold, and the sum of the two intensities is greater than a preset second intensity threshold, then the journal pair is determined to have a weight, and the first average of the two intensities is used as the weight of the journal pair. S502, construct a topology graph of the target domain based on all journals in the target domain and weighted journal pairs; the topology graph includes nodes and edges, a node is used to represent a journal, and the two nodes in a weighted journal pair are connected by an edge, the weight of the journal pair is the weight of the corresponding edge. S503, cluster all journals in the target domain according to the topology graph to obtain at least one first set.

[0073] In S501 of this application embodiment, for any journal pair, if both intensities in the intensity set of the journal pair are greater than a preset first intensity threshold, and the sum of the two intensities is greater than a preset second intensity threshold, it indicates that the two journals in the current journal pair cite each other too many times, and there is a suspicion of abnormal citation. Therefore, a weight is assigned to the journal pair, and the first average of the two intensities is used as the weight of the journal pair to provide corresponding data support for subsequent judgment of abnormal citation.

[0074] In S502 of this application embodiment, each journal in the target domain is treated as a node. For any pair of journals, it is determined whether the pair of journals has a weight. If the pair of journals has a weight, it indicates that there may be abnormal citations between the two journals in the pair. Therefore, the nodes representing the two journals are connected by an edge, and the weight of the pair of journals is used as the weight of the edge, thereby obtaining the topology graph of the target domain.

[0075] In S503 of this application embodiment, the topology graph can be clustered using the Louvain algorithm, and multiple journals connected by edges can be divided into a first set. The first set is a citation alliance that may have citation anomalies. In the subsequent citation anomaly detection, the first set divided based on the citation strength is used for detection, which greatly improves the detection accuracy.

[0076] In one example, the topology graph of the target domain is constructed in the following way: For any pair of journals (i,j), if any of the following conditions are satisfied, then an undirected edge e(i,j) is established between i and j: A[i][j]≥ θ1 and A[j][i]≥ θ1; (A[i][j] + A[j][i]) ≥ θ2; Wherein, θ1 is the first intensity threshold, with a value ranging from 0.3 to 0.8; θ2 is the second intensity threshold, with a value ranging from 0.8 to 1.5; the thresholds can be set according to the subject area through cross-validation or empirical values.

[0077] The weight of edge e(i, j) is w(i, j) = (A[i][j] + A[j][i]) / 2.

[0078] In the above scheme, by checking whether both of the two intensities in the intensity set are greater than a first intensity threshold and whether the sum of the two intensities is greater than a second intensity threshold, the scheme filters out journal pairs with excessively high citation intensity in the first set and assigns corresponding weights to these pairs. When constructing the topology graph of the target domain, each node is treated as a journal, and journal pairs with weights are connected by edges. This allows for a visual understanding of the journal pairs with excessively high citation intensity within the current target domain from the topology graph. Based on the topology graph, all journals within the target domain are clustered to obtain at least one first set. This achieves the division of journal pairs within the target domain by citation intensity, grouping journals with high citation intensity into a first set. The subsequent abnormal citation detection consortium provides the detection targets.

[0079] Based on the above embodiments, as an optional embodiment, a method for filtering the first set is also provided, such as... Figure 6 As shown, the specific content is as follows: S601, obtain the first total number of all journals in the first set, and determine the total number of journal pairs in the first set based on the first total number; S602, obtain the second total number of journal pairs in the first set whose two intensities are both greater than a preset first intensity threshold and whose sum of the two intensities is greater than a preset second intensity threshold; S603, the ratio of the second total number to the total logarithm is used as the degree of closeness among journals within the first set; S604, if it is determined that the first total quantity is less than a preset quantity threshold and the density is greater than a preset density threshold, then at least one of the first sets in the first set is retained.

[0080] In S601 of this application embodiment, after obtaining multiple first sets, the first sets can be further filtered by obtaining the first total number of journals in the first set and the total number of journal pairs in the first set, so as to select the first set that is more in line with abnormal citations to participate in subsequent abnormal detection, thereby improving the accuracy of abnormal citation detection.

[0081] In S602 of this application embodiment, for any journal pair, if both intensities in the intensity set of the journal pair are greater than a preset first intensity threshold, and the sum of the two intensities is greater than a preset second intensity threshold, it indicates that the two journals in the current journal pair cite each other too many times, and there is a suspicion of abnormal citation. The second total number of journal pairs suspected of abnormal citation in the first set is counted. In S603 of this application embodiment, since the second total number represents the number of journal pairs suspected of abnormal citation, the larger the second number, the more journal pairs with high intensity mutual citation within the first set there are. Therefore, by obtaining the ratio of the second total number to the total number of pairs, the closeness between journals within the first set can be obtained.

[0082] In S604 of this application embodiment, if it is determined that the first total number is less than a preset number threshold, it indicates that there are few journals in the first set, which belongs to the cooperation between small journals. If it is determined that the closeness is greater than the preset closeness threshold, it indicates that the citation relationship between journals is too close. The first set that meets the above conditions is likely to belong to a citation alliance with abnormal citations. Therefore, the first set is retained from at least one first set. If the above conditions are not met, the first set is removed from at least one first set. This achieves the filtering out of normal large academic cooperation communities by the above conditions, and retaining small-scale, highly close abnormal mutual citation journal groups.

[0083] In one example, the compactness of the first set is calculated as follows: :

[0084] in, For the second total quantity, This represents the total number of journals within the first set.

[0085] If the first set satisfies the following condition, it is marked as a candidate reference manipulation alliance, that is, the first set is retained in at least one of the first sets: |V(ck)| ≤ N_max, where N_max is the preset maximum alliance size threshold (quantity threshold), ranging from 3 to 15; (filtering out large, normal academic communities). Density(ck) ≥ ρ, where ρ is a preset density threshold, ranging from 0.4 to 0.8 (indicating that the internal network connections are sufficiently dense).

[0086] In the above scheme, by statistically analyzing the total number of all journal pairs in the first set and the second total number of journal pairs with excessively high intensity in the first set, the proportion of journal pairs with excessively high intensity to all journal pairs is obtained, thereby determining the degree of closeness between journals in the first set. At least one first set is retained, where the degree of closeness is greater than a preset closeness threshold and the first total number of journals is less than a preset number threshold. This achieves the filtering out of normal large academic collaboration communities and retains small-scale, high-closeness first sets to participate in subsequent abnormal citation detection, thus improving the accuracy and efficiency of abnormal detection.

[0087] Based on the above embodiments, as an optional embodiment, the first abstract text of the citing paper and the second abstract text of the cited paper are extracted; the cosine similarity between the first abstract text and the second abstract text is obtained; at least one first keyword in the citing paper and at least one second keyword in the cited paper are extracted; the number of identical keywords among the at least one first keyword and at least one second keyword is obtained; the total number of keyword types among the at least one first keyword and at least one second keyword is obtained; and the proportion of identical keywords to the total number of types is obtained. The cosine similarity and the proportion are weighted and summed according to a preset weight coefficient to obtain the semantic similarity.

[0088] In this embodiment of the application, both the citing paper and the cited paper are published in any journal within the first set, and there is a citation relationship between the citing paper and the cited paper.

[0089] In this embodiment, the first abstract text of the citing paper and the second abstract text of the cited paper are extracted, a TF-IDF vector is constructed, and the cosine similarity between the first abstract text and the second abstract text is calculated.

[0090] In this embodiment of the application, at least one first keyword of the citing paper and at least one second keyword of the cited paper are extracted. The intersection of the at least one first keyword and the at least one second keyword is taken to determine the number of identical keywords among the at least one first keyword and the at least one second keyword. The union of the at least one first keyword and the at least one second keyword is taken to determine the total number of keyword types of the at least one first keyword and the at least one second keyword. The Jaccard similarity coefficient is calculated based on the proportion of identical keywords to the total number of types.

[0091] In this embodiment of the application, the semantic similarity between the citing paper and the cited paper is obtained by weighting and summing the pre-set weight coefficients of cosine similarity and ratio based on the pre-set weight coefficients of cosine similarity and ratio.

[0092] In one example, the semantic similarity score Content_Score is determined using the following formula: Content_Score = α × Sim_TFIDF + (1-α) × Jaccard Jaccard= | Ka ∩ Kb | / | Ka ∪ Kb| Where α is the weight coefficient with a value of 0.75, Sim_TFIDF is the semantic similarity, Jaccard is the proportion of the number of identical keywords to the total number of categories, Ka is at least one primary keyword, and Kb is at least one secondary keyword.

[0093] In the above scheme, the cosine similarity of the paper abstract and the Jaccard coefficient of the keywords are introduced to quantify the content relevance between the cited paper and the cited paper. By introducing semantic similarity, the typical manipulation feature of "content-irrelevant but frequently cite each other" can be identified.

[0094] Based on the above embodiments, as an optional embodiment, a first anomaly probability is determined according to semantic similarity; the first anomaly probability is linearly negatively correlated with semantic similarity; an adjustment coefficient is determined according to the first number; the adjustment coefficient is monotonically increasing with the first number, and the increment gradually decreases; the first anomaly probability is adjusted by the adjustment coefficient to obtain the anomaly probability of the paper's citation relationship being abnormal.

[0095] In this embodiment of the application, the higher the semantic similarity between the citing paper and the cited paper, the higher the content relevance between the citing paper and the cited paper. Normally, the citing paper cites the cited paper based on the content relevance between the citing paper and the cited paper. Therefore, if the semantic similarity between the citing paper and the cited paper is low, it means that the citing paper cites the cited paper without any basis and is not a paper citation for academic purposes. Therefore, the lower the semantic similarity, the greater the probability of abnormal citation between the paper pairs. Therefore, the first abnormal probability of the paper pair can be determined based on the magnitude of semantic similarity.

[0096] In this embodiment of the application, the first number refers to the number of times the cited paper is cited by all papers published in journals within the first set. Therefore, the more first numbers there are, the greater the citation influence of the cited paper within the first set, and the higher the probability of abnormal citations. Thus, the adjustment coefficient of the first abnormal probability can be determined based on the first number.

[0097] In this embodiment of the application, since the smaller the semantic similarity, the greater the corresponding first anomaly probability, and the more inputs in the first time, the greater the corresponding adjustment coefficient, the first anomaly probability can be adjusted based on the adjustment coefficient to obtain the anomaly probability of the paper's abnormal citation relationship.

[0098] In one example, the anomalous probability of the paper pair (pa, pb) is calculated using the following formula. :

[0099] in, This represents the semantic similarity between the papers and (pa, pb). This represents the first number of times that the cited paper pb is cited by all papers published in the first set of journals. Here, log() is the natural logarithm, and the increment operation ensures that the logarithm result is 0 when the citation count is 0.

[0100] In the above scheme, semantic similarity is used to determine whether there is a reasonable citation motivation (content relevance) between the citing paper and the cited paper. The lower the semantic similarity, the lower the content relevance, and the less motivation the citing paper has to cite the cited paper. Therefore, the first anomaly probability is determined based on semantic similarity. The citation influence of the cited paper in the first set can be determined by the first number of citations. Therefore, the anomaly probability is calculated by combining the number of citations in the first set, which simultaneously reflects the two dimensions of "why cite" and "citation influence". This can more accurately identify mutual citations without substantial content relevance made for non-academic motivations (such as increasing impact factor) and improve the accuracy of anomaly citation detection.

[0101] Based on the above embodiments, as an optional embodiment, the second mean of the abnormal probability of all paper pairs with citation relationships in the first set is obtained; for each paper pair with citation relationships, if the abnormal probability of the paper pair is greater than a preset abnormal probability threshold, the paper pair is regarded as an abnormal paper pair; the third total number of all abnormal paper pairs in the first set and the fourth total number of all paper pairs in the first set are obtained; if the second mean is greater than a preset mean threshold, and the ratio of the third total number to the fourth total number is greater than a preset ratio threshold, it is determined that there are abnormal citations in the first set.

[0102] In this embodiment of the application, for each first set, the anomalous probability of all paper pairs with citation relationships within the first set is calculated, and the average of all anomalous probabilities is obtained to obtain a second mean, which can characterize the overall anomalous probability of the first set.

[0103] In this embodiment of the application, for each paper pair with a citation relationship, if the probability of anomaly in the paper pair is greater than a preset probability threshold, it indicates that the current paper pair has an abnormal citation. Therefore, the paper pair is regarded as an abnormal paper pair. The third total number of all abnormal paper pairs in the first set and the fourth total number of all paper pairs in the first set are counted, and the ratio of the third total number to the fourth total number is obtained. The proportion of abnormal paper pairs in the first set can be determined.

[0104] In this embodiment of the application, if the second mean is determined to be greater than the preset mean threshold, it indicates that the overall probability of anomalies in the first set is high and there is a high possibility of abnormal citations. If the ratio of the third total number to the fourth total number is determined to be greater than the preset proportion threshold, it indicates that there is a large proportion of paper pairs with abnormal citations in the first set. When both the overall probability of anomalies is high and the proportion of abnormal paper pairs is high, it indicates that the first set is a citation consortium with abnormal citations.

[0105] In one example, the first set containing abnormal references is filtered in the following way: For the first set ck, the second mean is calculated using the following formula. The ratio of the total number of third and abnormal papers :

[0106]

[0107] Where P is the total number of paper pairs in the first set. Let γ be the anomaly probability of the paper pair (pa, pb), and let γ be the anomaly probability threshold, ranging from 0.5 to 1.5.

[0108] For the first set ck, if the following conditions are met simultaneously, then the first set is determined to be a reference manipulation alliance with a reference exception: Avg_D(ck) ≥ μ, where μ is a preset mean threshold, ranging from 0.3 to 1.0; Phigh(ck) ≥ τ, where τ is a preset proportional threshold, ranging from 0.3 to 0.7.

[0109] It should be noted that the above thresholds can all be set through cross-validation or empirical ranges, without the need for manually customizing feature rules for each discipline or database, thus exhibiting good domain adaptability and transferability.

[0110] In the above scheme, the overall assessment of abnormal citations in the first set is achieved by obtaining the second mean of the abnormal probability of paper pairs with citation relationships within the first set. The proportion of abnormal papers in the first set is determined by obtaining the proportion of abnormal paper pairs in the first set to all paper pairs. When the second mean is greater than the mean threshold and the proportion is greater than the proportion threshold, it indicates that there are abnormal citations in the current first set. This achieves the assessment of abnormal citations in the first set from two dimensions, improving the accuracy of abnormal citation detection.

[0111] Based on the above embodiments, as an optional embodiment, after determining that the first set has abnormal references, at least one of the following is output: The names of the journals in the first set; At least one paper pair with the highest anomalous probability in the first set; The topological graph corresponding to the first set; The topology graph includes nodes and edges. A node represents a journal in the first set. The nodes representing the two journals in a pair of journals with anomalous paper pairs in the first set are connected by edges.

[0112] In this embodiment of the application, if it is determined that there are abnormal citations in the first set, the names of each journal in the first set are output to display the journals that currently form the citation consortium.

[0113] In this embodiment of the application, at least one paper pair with the highest probability of anomaly is obtained from the first set, and the paper pair is presented as interpretable evidence that the first set is a citation consortium, using the text portion cited in the citing paper and the cited portion in the cited paper.

[0114] In this embodiment of the application, the calculated second mean of the first set and the proportion of abnormal paper pairs can also be output as evidence of interpretability.

[0115] In this embodiment of the application, a topology graph of the first set can also be output. The topology graph of the first set is used to visualize the mutual citation relationship between journals in the first set. Each node represents the first journal in the first set. The two nodes represented by the journal pairs with abnormal paper pairs are connected by an edge, which serves as the interpretability basis for the first set as a citation alliance with abnormal citations.

[0116] In the above scheme, by outputting the names of each journal in the first set, the second mean of the first set, the proportion of abnormal paper pairs, the paper pairs with the highest abnormal probability in the first set, and the topology graph of the first set, it not only detects citation alliances with abnormal citations, but also provides corresponding specific evidence to support the detection results, making the detection results highly interpretable and facilitating manual review and subsequent processing.

[0117] The abnormal citation detection method provided in this application, through the abnormal probability of a first set, does not require manually setting complex rules or relying on a few explicit abnormal indicators. It can automatically identify potential citation manipulation alliances from massive amounts of literature metadata, making it suitable for large-scale, interdisciplinary research integrity monitoring. By determining the expected citation count based on the actual citation count and calculating the strength set of each journal pair, it can effectively distinguish between normal academic citations and statistically abnormal co-citation behavior. At the same time, by clustering the density with the first set, it filters large normal academic communities, reducing the false alarm rate. It can identify bidirectional, high-density "citation clustering" behavior formed between multiple journals, making up for the shortcomings of related methods in detecting collaborative manipulation.

[0118] This application provides an abnormal reference detection device, such as... Figure 7 As shown, the anomaly detection device 70 may include: an acquisition module 701, a clustering module 702, a determination module 703, and a judgment module 704.

[0119] Specifically, the acquisition module 701 is used to acquire the citation count between each journal pair in the target field. For each journal pair, the intensity set of the journal pair is determined based on the citation count between the journal pairs and the total citation count between all journals. The intensity set is used to characterize the intensity of mutual citation between the papers published in the two journals included in the journal pair. The intensity set includes the intensity of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal. Clustering module 702 is used to cluster journals in the target domain according to the intensity set of each journal pair to obtain at least one first set; each first set includes at least one journal. The determination module 703 is used to, for each first set, for each paper pair in the first set that has a citation relationship, obtain the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by papers in all journals in the first set, and determine the abnormal probability of the citation relationship of the paper pair based on the semantic similarity and the first number of citations. The judgment module 704 is used to determine whether there are abnormal citations in each first set based on the abnormal probability of all paper pairs with citation relationships in the first set.

[0120] The abnormal citation detection device provided in this application obtains the citation count between each journal pair in the target field. For each journal pair, it determines the intensity set of the journal pair based on the citation count between the journal pairs and the total citation count between all journal pairs. This realizes the quantification of the intensity of mutual citation between the papers published in the two journals included in each journal pair by using the citation count between the two journals in the journal pair and the total citation count between all journal pairs.

[0121] By clustering journals in the target field according to the intensity set of each journal pair, at least one first set including at least one journal is obtained. This realizes the ability to group journals with high mutual citation intensity into a first set based on the mutual citation intensity between journal pairs, so as to perform abnormal citation detection on each first set in the future. For each first set, for each paper pair with a citation relationship in the first set, the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by all journals in the first set are obtained. The abnormal probability of the citation relationship of the paper pair is determined based on the semantic similarity and the first number of citations. The content relevance between the citing paper and the cited paper is quantified by the semantic similarity. Combined with the content relevance and the first number of times the citing paper is cited by all journals, the abnormal probability of the citation relationship of the paper pair is determined. This enables more accurate identification of citation behavior in paper pairs that is not motivated by academic research. For each set, the presence of aberrant citations in the first set is determined by the aberration probability of all paper pairs with citation relationships within the first set. By combining the aberration probabilities of all paper pairs with citation relationships within the first set, the presence of aberrant citations in the first set as a whole is determined. Thus, the determination of whether the first set is a citation manipulation alliance with citation manipulation behavior can be made based on the judgment result. This solves the problem of distinguishing between reasonable academic cooperation (such as a group of journals with similar disciplines) and maliciously formed citation manipulation alliances.

[0122] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0123] This application provides an electronic device (computer device / equipment / system) including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of an abnormal citation detection method. Compared with related technologies, this method solves the problem of difficulty in distinguishing between reasonable academic collaborations (such as a group of journals with similar disciplines) and maliciously formed citation manipulation alliances.

[0124] In one alternative embodiment, an electronic device is provided, such as Figure 8 As shown, Figure 8 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this application.

[0125] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0126] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0127] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0128] The memory 4003 stores computer programs that execute embodiments of this application, and its execution is controlled by the processor 4001. The processor 4001 executes the computer programs stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0129] The electronic device package may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 8 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0130] This application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program can implement the steps and corresponding content of the aforementioned method embodiments. Compared with the prior art, this solves the problem of difficulty in distinguishing between legitimate academic collaborations (such as a group of journals with similar disciplines) and maliciously formed citation manipulation alliances.

[0131] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium, a computer-readable medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0132] This application also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments. Compared with the prior art, it can achieve: It solves the problem of distinguishing between legitimate academic collaborations (such as a group of journals with similar disciplines) and malicious citation manipulation alliances.

[0133] The terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the illustrations or text descriptions.

[0134] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0135] The above are only optional implementation methods for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application, without departing from the technical concept of this application, also fall within the protection scope of the embodiments of this application.

Claims

1. A method for detecting abnormal references, characterized in that, include: The citation counts between each journal pair within the target field are obtained. For each journal pair, the intensity set of the journal pair is determined based on the citation counts between the journal pairs and the total citation counts between all journals. The intensity set is used to characterize the intensity of mutual citations between papers published in the two journals included in the journal pair. The intensity set includes the intensity of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal. Based on the intensity set of each journal pair, the journals in the target domain are clustered to obtain at least one first set; each first set includes at least one journal. For each first set, for each paper pair in the first set that has a citation relationship, obtain the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by papers in all journals in the first set, and determine the abnormal probability of the citation relationship of the paper pair based on the semantic similarity and the first number of times. For each first set, determine whether there are anomalous citations in the first set based on the anomalous probability of all paper pairs with citation relationships within the first set.

2. The method according to claim 1, characterized in that, The strength of the citing journal's citation of the cited journal is determined by the following method: The second number of times the citing journal cites the cited journal, the first total number of times the citing journal cites all journals in the target field, the second total number of times the cited journal is cited by each journal in the target field, and the third total number of times each journal in the target field cites the other two journals. The expected citation count is determined based on the first total number of citations, the second total number of citations, and the third total number of citations; the expected citation count refers to the expected number of times the citing journal cites the cited journal within the third total number of citations. The intensity of the citing journal's citation of the cited journal is determined based on the expected number of times, the second number of times, and the first total number of times.

3. The method according to claim 2, characterized in that, The step of determining the expected citation count based on the first total count, the second total count, and the third total count includes: Based on the ratio of the first total number of citations to the third total number of citations, a first probability is determined in the third total number of citations that the citing journal cites all journals in the target field; Based on the ratio of the second total number of citations to the third total number of citations, a second probability is determined that, in the third total number of citations, the cited journal is cited by other journals in the target field. Based on the first probability and the second probability, determine the third probability that the citing journal will cite the cited journal in the third total number of citations; The expected number of references is determined based on the third probability and the third total number of references.

4. The method according to claim 2, characterized in that, Determining the strength of the citation of the cited journal by the citing journal based on the expected number of citations, the second number of citations, and the first total number of citations includes: Based on the ratio between the second number of citations and the first total number of citations, a fourth probability is determined of the citing journal citing the cited journal in the first total number of citations. The degree of difference between the actual number of times the citing journal cites the cited journal and the expected number of times is determined based on the ratio between the second number of citations and the expected number of citations. The strength of the citing journal's citation of the cited journal is determined based on the degree of difference and the fourth probability.

5. The method according to claim 2, characterized in that, The step of clustering journals within the target domain based on the intensity set of each journal pair to obtain at least one first set includes: For each journal pair, if both intensities in the intensity set of the journal pair are greater than a preset first intensity threshold, and the sum of the two intensities is greater than a preset second intensity threshold, then the journal pair is determined to have a weight, and the first mean of the two intensities is used as the weight of the journal pair. Based on all journals in the target domain and weighted journal pairs, a topology graph of the target domain is constructed. The topology graph includes nodes and edges. A node represents a journal. The two nodes in a weighted journal pair are connected by an edge. The weight of the journal pair is the weight of the corresponding edge. Based on the topology graph, all journals in the target domain are clustered to obtain at least one first set.

6. The method according to claim 1, characterized in that, After obtaining the first set, the process also includes: Obtain the first total number of all journals in the first set, and determine the total number of journal pairs in the first set based on the first total number; Obtain the second total number of journal pairs in the first set whose two intensities are both greater than a preset first intensity threshold and whose sum of the two intensities is greater than a preset second intensity threshold; The ratio of the second total number to the total logarithm is used as the degree of closeness among journals within the first set; If it is determined that the first total quantity is less than a preset quantity threshold, and the density is greater than a preset density threshold, then the first set in the at least one first set is retained.

7. The method according to claim 1, characterized in that, The step of obtaining the semantic similarity between the citing paper and the cited paper in the paper pair includes: Extract the first abstract text of the citing paper and the second abstract text of the cited paper; Obtain the cosine similarity between the first summary text and the second summary text; Extract at least one first keyword from the citing paper and at least one second keyword from the cited paper; Obtain the number of identical keywords among the at least one first keyword and the at least one second keyword; obtain the total number of keyword types among the at least one first keyword and the at least one second keyword; and obtain the proportion of the number of identical keywords to the total number of types. The semantic similarity is obtained by weighting and summing the cosine similarity and the ratio according to preset weight coefficients.

8. The method according to claim 1, characterized in that, Determining the probability of anomalies in the citation relationships of the paper based on the semantic similarity and the first count includes: A first anomaly probability is determined based on the semantic similarity; the first anomaly probability is linearly negatively correlated with the semantic similarity. An adjustment coefficient is determined based on the first number of times; the adjustment coefficient is monotonically increasing with the first number of times, and the increment gradually decreases; The first anomaly probability is adjusted by the adjustment coefficient to obtain the anomaly probability of the paper's citation relationship anomaly.

9. The method according to claim 1, characterized in that, The step of determining whether there are anomalous citations in the first set based on the anomalous probability of all paper pairs with citation relationships within the first set includes: Obtain the second mean of the anomaly probability of all paper pairs with citation relationships within the first set; For each paper pair with a citation relationship, if the probability of an anomaly in the paper pair is greater than a preset anomaly probability threshold, then the paper pair is considered an anomaly paper pair. Obtain the third total number of all abnormal paper pairs in the first set and the fourth total number of all paper pairs in the first set; If the second mean is greater than a preset mean threshold, and the ratio of the third total quantity to the fourth total quantity is greater than a preset ratio threshold, then it is determined that the first set has abnormal references.

10. The method according to claim 9, characterized in that, After determining that the first set contains abnormal references, the process further includes: Output at least one of the following: The names of the journals in the first set; At least one paper pair with the highest probability of anomaly in the first set; The topological graph corresponding to the first set; The topology graph includes nodes and edges. A node represents a journal within the first set, and the nodes representing the two journals in a pair of journals with anomalous paper pairs within the first set are connected by the edges.

11. An anomalous reference detection device, characterized in that, include: The acquisition module is used to acquire the citation count between each journal pair within the target field. For each journal pair, the intensity set of the journal pair is determined based on the citation count between the journal pair and the total citation count between all journals. The intensity set is used to characterize the intensity of mutual citation between papers published in the two journals included in the journal pair. The intensity set includes the intensity of the citation of the cited journal by the citing journal when each journal in the journal pair is the citing journal and the other journal is the cited journal. The clustering module is used to cluster journals in the target domain according to the intensity set of each journal pair to obtain at least one first set; each first set includes at least one journal. The determination module is used to, for each first set, for each paper pair in the first set that has a citation relationship, obtain the semantic similarity between the citing paper and the cited paper in the paper pair and the first number of times the cited paper is cited by papers in all journals in the first set, and determine the abnormal probability of the citation relationship of the paper pair based on the semantic similarity and the first number of times. The judgment module is used to determine whether there are abnormal citations in each first set based on the abnormal probability of all paper pairs with citation relationships in the first set.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-10.

Images