A keyword reverse propagation algorithm based on citation network structure
By using a keyword backpropagation algorithm based on citation network structure and optimizing keyword weights with a spring charge model and a neural network model, the problem of poor visualization effect of citation network in traditional algorithms is solved, and a clearer visualization effect is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- UNICLOUD TECH CO LTD
- Filing Date
- 2023-05-12
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional document clustering algorithms struggle to maintain the complete structure of the citation network, resulting in poor visualization performance.
A keyword backpropagation algorithm based on citation network structure is adopted. By combining the spring charge model and neural network model with force-directed layout and iterative calculation, the keyword weights are adjusted to optimize the layout of the citation network.
It achieves a clearer visualization of citation networks while preserving the integrity and similarity of text clustering.
Smart Images
Figure CN116680397B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of citation networks, and in particular relates to a keyword backpropagation algorithm based on citation network structure. Background Technology
[0002] A citation network is a collection of citation and reference relationships between documents. These documents include various forms such as scientific journals, patent documents, conference proceedings, scientific reports, and dissertations, effectively describing the development of scientific fields and the relationships between disciplines. With the development of modern information technology, the number of publications has increased rapidly, and citation networks have formed a massive and complex network system, attracting increasing attention. Because citation networks encompass research results from multiple fields, representing an important knowledge repository in academic research, they have become an important medium for research.
[0003] Cluster analysis, also known as group analysis, is a statistical analysis method for studying the classification of samples or indicators, and it is also an important algorithm in data mining. Cluster analysis consists of several patterns, typically a vector of measurements or a point in a multidimensional space. Cluster analysis is based on similarity; patterns within a cluster are more similar than patterns in different clusters.
[0004] Text-based clustering algorithms typically input vectorized corpora and then calculate the similarity between different corpora to determine the similarity between texts. Traditional document clustering algorithms usually use document abstracts and keywords as input, which makes it difficult to guarantee a complete citation network structure. Therefore, in visualization, this can lead to cross-citation between documents in different clusters, resulting in poor visualization effects. Summary of the Invention
[0005] In view of this, the present invention aims to propose a keyword backpropagation algorithm based on citation network structure to obtain the complete citation network structure, thereby achieving better visualization results.
[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:
[0007] A keyword backpropagation algorithm based on a citation network structure includes the following steps:
[0008] S1: Establish a spring charge model and perform force-directed layout processing to create a force-directed layout diagram;
[0009] S2: Using the back propagation algorithm, a keyword propagation model is established to obtain a comparison curve of keyword weight changes;
[0010] S3: Using the force-directed layout diagram established in step S1, construct the citation network model;
[0011] S4: Iteratively calculate the force-directed layout diagram until the energy state in the force-directed layout diagram reaches the minimum value;
[0012] S5: Use the iterative calculation method in step S4 to replace the iterative calculation method in the citation network model established in step S3, and perform iterative calculation;
[0013] S6: While iteratively calculating the citation network in step S5, the keyword weights in the citation network model are adjusted using the keyword weight change comparison curve obtained in step S2.
[0014] S7: The converged citation network layout diagram is obtained through steps S1 to S6.
[0015] Furthermore, the specific process of establishing the force-directed layout diagram in step S1 is as follows:
[0016] S101: Treat each energetic discharge particle in the spring charge model as a node in the force-directed layout;
[0017] S102: Based on the Coulomb force and Hooke gravity between particles, the correlation of the forces between two particles is calculated. The correlation of the forces is used as the connection between two nodes to obtain the edges in the force-oriented layout.
[0018] S103: Use the nodes and edges obtained in steps S101 and S102 to create a force-guided layout diagram.
[0019] Furthermore, the process of establishing the related word propagation model in step S2 is as follows:
[0020] S201: Establish a neural network model, input keyword data into the neural network for computation, calculate the output of each neuron, and generate the final output result;
[0021] S202: Calculate the error between the actual output and the target output, substitute the error value into the error function, and calculate the derivative of the keyword weight with respect to the bias.
[0022] S203: Pass the derivatives of the keyword weights and biases back to the preceding layers of the network to update the keyword weights and biases in the preceding layers.
[0023] S204: Using the original keyword weights and biases, as well as the updated keyword weights and biases, arrange them to obtain the actual change curve and the target change curve, and then obtain the keyword change comparison curve.
[0024] Furthermore, the steps for establishing the citation network model in step S3 are as follows:
[0025] S301: Collect a certain amount of academic literature data and convert the academic literature data into an interactive data format;
[0026] S302: Extract the necessary information from the literature data as keywords;
[0027] S303: Construct a directed graph where each node represents an academic paper and each edge represents a link to other papers in the paper.
[0028] S304: Use clustering algorithms to divide the network into subgroups and identify core nodes and key documents;
[0029] S305: Iterative calculation yields the citation network layout diagram.
[0030] Furthermore, the iterative calculation process for the force-directed layout diagram in step S4 is as follows:
[0031] S401: Input the number of iterations, data nodes, and the edges corresponding to the data nodes;
[0032] S402: In each iteration, all data nodes are traversed, the Coulomb repulsion function is called, and recursive calculations are performed to obtain the electrostatic repulsion between all data nodes.
[0033] S403: Traverse the edges corresponding to all data nodes, call the Hooke gravity function, and recursively calculate the electrostatic repulsion data obtained in step S402 to obtain new electrostatic repulsion data.
[0034] S404: Traverse all data nodes again and call the position function to update the position of the data nodes. The parameters of the position function include the current coordinates of the data nodes and the new electrostatic repulsion data obtained in step S403.
[0035] S405: Returns the updated data node to obtain location data.
[0036] Compared with existing technologies, the keyword backpropagation algorithm based on citation network structure described in this invention has the following advantages:
[0037] The keyword backpropagation algorithm based on citation network structure described in this invention takes the entire network as the main body, selects the keywords owned by the cited documents and propagates them backward along the network with a certain probability from the cited documents to the citing documents. While retaining the idea of text clustering, it increases the weight of keywords and obtains clear and visualized data effects through force-directed layout. Attached Figure Description
[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:
[0039] Figure 1 This is a schematic diagram of a keyword backpropagation algorithm based on a citation network structure as described in an embodiment of the present invention;
[0040] Figure 2 This is a schematic diagram of a visualized network obtained through an existing citation network, as described in an embodiment of the present invention.
[0041] Figure 3 This is a schematic diagram of a visual network obtained by using the improved citation network of the present invention, as described in an embodiment of the present invention. Detailed Implementation
[0042] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0043] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0044] A keyword backpropagation algorithm based on a citation network structure includes the following steps:
[0045] S1: Establish a spring charge model and perform force-directed layout processing to create a force-directed layout diagram;
[0046] S2: Using the back propagation algorithm, a keyword propagation model is established to obtain a comparison curve of keyword weight changes;
[0047] S3: Using the force-directed layout diagram established in step S1, construct the citation network model;
[0048] S4: Iteratively calculate the force-directed layout diagram until the energy state in the force-directed layout diagram reaches the minimum value;
[0049] S5: Use the iterative calculation method in step S4 to replace the iterative calculation method in the citation network model established in step S3, and perform iterative calculation;
[0050] S6: While iteratively calculating the citation network in step S5, the keyword weights in the citation network model are adjusted using the keyword weight change comparison curve obtained in step S2.
[0051] S7: The converged citation network layout diagram is obtained through steps S1 to S6.
[0052] The specific process of establishing the force-directed layout diagram in step S1 is as follows:
[0053] S101: Treat each energetic discharge particle in the spring charge model as a node in the force-directed layout;
[0054] S102: Based on the Coulomb force between particles and Hooke's gravity F s =k s (x-x0) is used to calculate the correlation of the forces between the two particles. The correlation of the forces is used as the connection between the two nodes to obtain the edges in the force-directed layout.
[0055] S103: Use the nodes and edges obtained in steps S101 and S102 to create a force-guided layout diagram.
[0056] Under the influence of Coulomb repulsion and Hooke attraction, particles are displaced in the direction of the specified resultant force. From an overall perspective, the particles continuously shift from the initial random disordered state to a stable final state of equilibrium and order. At the same time, the energy of the entire physical system is constantly being consumed. After multiple iterations, the particles no longer undergo relative displacement. At this point, the entire system reaches a stable state, which is the state of minimum energy.
[0057] The process of establishing the related word propagation model in step S2 is as follows:
[0058] S201: Establish a neural network model, input keyword data into the neural network for computation, calculate the output of each neuron, and generate the final output result;
[0059] S202: Calculate the error between the actual output and the target output, substitute the error value into the error function, and calculate the derivative of the keyword weight with respect to the bias.
[0060] S203: Pass the derivatives of the keyword weights and biases back to the preceding layers of the network to update the keyword weights and biases in the preceding layers.
[0061] S204: Using the original keyword weights and biases, as well as the updated keyword weights and biases, arrange them to obtain the actual change curve and the target change curve, and then obtain the keyword change comparison curve.
[0062] For each document, there are n keywords k1, k2, k3, ..., k n Each keyword k i (0≤i≤n) corresponds to a positive integer w i , representing the total number of times this keyword appears in all documents, is used during dissemination. Each keyword k... i The probability of being backpropagated is: To avoid losing the probability of this propagation, by adjusting the selection of σ, we can avoid the situation where the entire local network propagates the same keyword. This prevents the entire network from overfitting the original citation network structure and losing basic text similarity details.
[0063] Error function:
[0064] Where p(x) is the actual output and q(x) is the target output.
[0065] The steps for establishing the citation network model in step S3 are as follows:
[0066] S301: Collect a certain amount of academic literature data and convert the academic literature data into an interactive data format (e.g., XML or JSON);
[0067] S302: Extract necessary information from literature data as keywords, such as title, abstract, author, publication time, etc.
[0068] S303: Construct a directed graph in which each node represents an academic paper and each edge represents a link to other papers in the paper. If a paper is cited multiple times, increase the weight of the corresponding node to reflect its influence in the academic community.
[0069] S304: Use clustering algorithms to divide the network into subgroups and identify core nodes and key documents;
[0070] S305: Iterative calculation yields the citation network layout diagram.
[0071] The iterative calculation process for the force-directed layout diagram in step S4 is as follows:
[0072] S401: Input the number of iterations, data nodes, and the edges corresponding to the data nodes;
[0073] S402: In each iteration, all data nodes are traversed, the Coulomb repulsion function is called, and recursive calculations are performed to obtain the electrostatic repulsion between all data nodes.
[0074] S403: Traverse the edges corresponding to all data nodes, call the Hooke gravity function, and recursively calculate the electrostatic repulsion data obtained in step S402 to obtain new electrostatic repulsion data.
[0075] S404: Traverse all data nodes again and call the position function to update the position of the data nodes. The parameters of the position function include the current coordinates of the data nodes and the new electrostatic repulsion data obtained in step S403.
[0076] S405: Returns the updated data node to obtain location data.
[0077] The pseudocode for iterative calculation is as follows:
[0078] Algorithm:(k,u,e)
[0079] Input: number of iterations k, all nodes u, all edges e
[0080] Output: The position of each node
[0081]
[0082] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0083] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the division of units described above is merely a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The aforementioned units may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention according to actual needs.
[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.
[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A keyword backpropagation algorithm based on citation network structure, characterized in that: Includes the following steps: S1: Establish a spring charge model and perform force-directed layout processing to create a force-directed layout diagram; S2: Using the back propagation algorithm, a keyword propagation model is established to obtain a comparison curve of keyword weight changes; S3: Using the force-directed layout diagram established in step S1, construct the citation network model; S4: Iteratively calculate the force-directed layout diagram until the energy state in the force-directed layout diagram reaches the minimum value; S5: Use the iterative calculation method in step S4 to replace the iterative calculation method in the citation network model established in step S3, and perform iterative calculation; S6: While iteratively calculating the citation network in step S5, the keyword weights in the citation network model are adjusted using the keyword weight change comparison curve obtained in step S2. S7: The converged citation network layout diagram obtained through steps S1 to S6; The process of establishing the related word propagation model in step S2 is as follows: S201: Establish a neural network model, input keyword data into the neural network for computation, calculate the output of each neuron, and generate the final output result; S202: Calculate the error between the actual output and the target output, substitute the error value into the error function, and calculate the derivative of the keyword weight with respect to the bias. S203: Pass the derivatives of the keyword weights and biases back to the preceding layers of the network to update the keyword weights and biases in the preceding layers. S204: Using the original keyword weights and biases, as well as the updated keyword weights and biases, arrange them to obtain the actual change curve and the target change curve, and then obtain the keyword change comparison curve.
2. The keyword backpropagation algorithm based on citation network structure according to claim 1, characterized in that: The specific process of establishing the force-directed layout diagram in step S1 is as follows: S101: Treat each energetic discharge particle in the spring charge model as a node in the force-directed layout; S102: Based on the Coulomb force and Hooke gravity between particles, the correlation of the forces between two particles is calculated. The correlation of the forces is used as the connection between two nodes to obtain the edges in the force-oriented layout. S103: Use the nodes and edges obtained in steps S101 and S102 to create a force-guided layout diagram.
3. The keyword backpropagation algorithm based on citation network structure according to claim 1, characterized in that: The steps for establishing the citation network model in step S3 are as follows: S301: Collect a certain amount of academic literature data and convert the academic literature data into an interactive data format; S302: Extract the necessary information from the literature data as keywords; S303: Construct a directed graph where each node represents an academic paper and each edge represents a link between the academic paper represented by the corresponding node and other papers. S304: Use clustering algorithms to divide the network into subgroups and identify core nodes and key documents; S305: Iterative calculation yields the citation network layout diagram.
4. The keyword backpropagation algorithm based on citation network structure according to claim 1, characterized in that: The iterative calculation process for the force-directed layout diagram in step S4 is as follows: S401: Input the number of iterations, data nodes, and the edges corresponding to the data nodes; S402: In each iteration, all data nodes are traversed, the Coulomb repulsion function is called, and recursive calculations are performed to obtain the electrostatic repulsion between all data nodes. S403: Traverse the edges corresponding to all data nodes, call the Hooke gravity function, and recursively calculate the electrostatic repulsion data obtained in step S402 to obtain new electrostatic repulsion data. S404: Traverse all data nodes again and call the position function to update the position of the data nodes. The parameters of the position function include the current coordinates of the data nodes and the new electrostatic repulsion data obtained in step S403. S405: Returns the updated data node to obtain location data.