A knowledge graph driven multi-dimensional data acquisition and auxiliary decision system
The knowledge graph-driven multidimensional data acquisition system solves the problems of low efficiency in integrating multi-source heterogeneous data, insufficient accuracy in sentiment analysis, and lagging knowledge graph updates. It enables real-time association of multi-source data and efficient decision support, thereby improving data processing efficiency and decision accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TIBET LANSA ZHIHUI TECHNOLOGY CO LTD
- Filing Date
- 2026-03-20
- Publication Date
- 2026-07-31
AI Technical Summary
Existing technologies are inefficient in integrating multi-source heterogeneous data, lack sufficient accuracy in sentiment analysis, are lagging in knowledge graph updates, and have poor interpretability in decision support, resulting in insufficient real-time decision response capabilities and increased decision costs.
The knowledge graph-driven multidimensional data acquisition system achieves deep integration of heterogeneous data and real-time decision support through the combination of data acquisition layer, knowledge graph layer, analysis engine layer, decision support layer and user interaction layer, and uses deep learning to improve the accuracy of sentiment mining and the reliability of decision reasoning.
It achieves second-level association of multi-source heterogeneous data, improves the accuracy of sentiment analysis, reduces the knowledge graph update latency to the minute level, and enhances decision transparency and credibility, meeting the needs of minute-level decision response.
Smart Images
Figure CN122489254A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer application technology, specifically relating to a knowledge graph-driven multidimensional data acquisition and decision support system. Background Technology
[0002] With the rapid development of big data technology, data collection and decision support systems have been widely applied in various fields of society and the economy. However, the existing technological system still faces significant bottlenecks. In terms of integrating multi-source heterogeneous data, traditional systems often employ a siloed architecture, making it difficult to achieve semantic-level correlation between different modalities such as social media text, IoT sensor data, and enterprise business data. Data correlation efficiency often remains at the hourly level, severely restricting real-time decision-making response capabilities. While sentiment analysis technology has achieved basic polarity judgment, its accuracy in identifying implicit stances is generally low, especially when dealing with rhetorical expressions such as sarcasm, where the error rate increases significantly, failing to capture subtle changes in public opinion. Knowledge graph applications are mostly limited to static construction models, with knowledge update cycles lasting weeks or even months. For example, a recommendation system on an e-commerce platform experienced a decline in recommendation accuracy due to its failure to promptly incorporate new product category relationships. Decision support systems generally suffer from a "black box" dilemma. For instance, while a financial risk control model can output risk scores, it cannot explain the key factors underlying the scores. Regulatory agencies therefore require additional manual review processes, which actually increases decision-making costs.
[0003] Existing technologies suffer from difficulties in integrating multi-source data, reliance on a single data source, incompatibility of heterogeneous data formats, and low correlation efficiency. Furthermore, they lack precision in sentiment analysis, identifying only explicit sentiments while ignoring contextual dependence and implicit stances. Additionally, the static nature of knowledge graphs leads to delayed knowledge updates, failing to reflect real-time data changes. Moreover, existing technologies exhibit poor interpretability in decision support due to their black-box model outputs and lack causal reasoning links.
[0004] Therefore, this invention provides a knowledge graph-driven multidimensional data acquisition and decision support system. This system organically integrates multiple technical branches such as natural language processing, machine learning, knowledge engineering, and sentiment computing to form a full-link intelligent processing framework covering data perception, semantic understanding, sentiment analysis, and decision support. Summary of the Invention
[0005] To address the problems raised in the background technology, the purpose of this invention is to provide a knowledge graph-driven multidimensional data acquisition and decision support system. Its core lies in breaking through the single-modal limitations of traditional data acquisition and decision support systems, achieving deep integration of heterogeneous data sources through the semantic association capabilities of knowledge graphs, and simultaneously improving the accuracy of sentiment mining and the reliability of decision reasoning in complex scenarios with the help of deep learning technology.
[0006] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:
[0007] A knowledge graph-driven multidimensional data acquisition and decision support system includes a data acquisition layer, a knowledge graph layer, an analysis engine layer, a decision support layer, and a user interaction layer. Each layer achieves data interaction through standardized interfaces.
[0008] The data acquisition layer is deployed on edge nodes and connected to the knowledge graph layer. It is configured with multiple protocol adaptation sub-modules and supports HTTP / HTTPS, MQTT and WebSocket access protocols.
[0009] The knowledge graph layer adopts a hybrid storage architecture, receives standardized data pushed by the data collection layer through a REST API interface, and provides data services to the analysis engine layer.
[0010] The analysis engine layer is deployed on GPU computing nodes, and calls knowledge graph layer data through the gRPC protocol to configure the GPU to achieve parallel computing.
[0011] The decision support layer pushes decision suggestions to the user interaction layer via the SOAP protocol;
[0012] The user interaction layer is deployed on a CDN network and uses the WebSocket protocol to achieve real-time data visualization.
[0013] Furthermore, the data acquisition layer includes a three-level filtering mechanism: filtering spam advertising content based on a rule base; identifying outliers using an isolated forest algorithm; and removing contradictory data using contextual semantic consistency verification.
[0014] It also includes a feature extraction module connected to the data acquisition layer. The feature extraction module customizes extraction schemes for different data types. Text data is represented by a fusion of TF-IDF and BERT embedding vectors. Time series data is extracted by wavelet transform to extract frequency domain features. Spatial data is converted into GeoHash encoding and coordinate normalization representation.
[0015] Furthermore, the analysis engine layer includes a sentiment analysis submodule, adopts a bidirectional encoder architecture, is pre-trained based on the BERT-base model and fine-tuned with a domain sentiment dictionary, introduces a context-aware attention mechanism, and supports explicit sentiment polarity judgment and implicit stance recognition.
[0016] Further specifying, the decision support layer constructs a dynamic reasoning framework based on an improved Bayesian network, and the formula for calculating node influence is:
[0017]
[0018] in, As an influence decay factor, Let O(u,t) represent the causal strength from u to v at time t, O(u,t) be the out-degree of node u, and S(v,t) be the sentiment tendency value of node v.
[0019] Furthermore, the decision support layer also includes a multi-hypothesis reasoning unit, which uses an evidence theory framework to calculate hypothesis credibility. The hypothesis credibility calculation formula is as follows:
[0020]
[0021] Among them, the conflict coefficient .
[0022] Furthermore, the decision support layer also includes a causal path visualization unit, which is used to calculate the influence of key nodes, using the following formula:
[0023]
[0024] in, Let S(u) be the damping coefficient, S(u) be the influence score of node u, L(u) be the out-degree of node u, and N be the total number of nodes in the network.
[0025] Furthermore, the knowledge graph layer adopts a partitioned storage strategy to control the query latency of hot entities to less than 8ms; the knowledge graph layer also adopts a distributed graph database cluster architecture and is configured with a master-slave replication mechanism to achieve automatic fault switching, controlling the data synchronization latency to less than 100ms.
[0026] A knowledge graph-driven multidimensional data acquisition and decision support method, wherein the multi-granularity knowledge graph construction method of the knowledge graph layer includes the following steps:
[0027] S1: Data access, receiving standardized feature vectors from the data acquisition layer;
[0028] S2: Entity recognition, using an improved bidirectional LSTM-CRF model to perform entity recognition and disambiguation on multi-source data;
[0029] S3: Relation extraction, which extracts relationships between entities by combining syntactic dependency parsing and remote supervision techniques;
[0030] S4: Dynamic update, introducing a time-series decay factor to achieve minute-level updates of the knowledge graph and calculate entity weights. The entity weight adjustment formula is:
[0031]
[0032] in, As the initial weights, The attenuation coefficient is... The value range is 0.012-0.058. Let be the correlation strength between entity e and event i at time t. Score the impact of the event.
[0033] Furthermore, the S4 dynamic update also includes a context-aware weight adjustment mechanism, which dynamically adjusts the knowledge weights using the health status parameter ct and the treatment stage parameter ht, with the calculation formula as follows:
[0034]
[0035] in, This is the entity importance coefficient. As a time-dependent decay factor, For health status parameters, The value range is 0-1. These are parameters for the diagnosis and treatment phase.
[0036] The beneficial effects of this invention are:
[0037] This invention achieves significant breakthroughs in key indicators such as data association efficiency, sentiment recognition accuracy, knowledge update speed, and system concurrency by setting up a data acquisition layer, a knowledge graph layer, an analysis engine layer, a decision support layer, and a user interaction layer.
[0038] In the processing of multi-source heterogeneous data, traditional systems use ETL offline mode, which results in data association taking up to hours. This invention shortens the association time to seconds through a distributed real-time processing architecture, which can meet the needs of minute-level decision response.
[0039] In terms of sentiment analysis capabilities, existing models generally have an accuracy rate of less than 60% in identifying implicit stances, and an error rate as high as 39% for complex expressions such as metaphors and irony. This system adopts a bidirectional encoder architecture and domain-adaptive fine-tuning technology to improve the implicit sentiment recognition rate, especially in the product review SARCASM recognition scenario, significantly reducing the risk of misjudgment.
[0040] Regarding the dynamic updating of knowledge graphs, traditional construction methods have an average update cycle of 14 days, resulting in an entity missing rate of up to 43% in sudden scenarios. This invention introduces an incremental time-series update mechanism and a decay coefficient λ to achieve dynamic weight adjustment, compressing the knowledge graph update delay to the minute level and ensuring the timeliness of entities and relationships.
[0041] Regarding the reliability of decision support, existing systems suffer from a decision suggestion adoption rate of only 62% due to the black-box model output. This invention improves decision accuracy through time-varying causal propagation networks and interpretable inference engines; it also supports causal path visualization, with each path containing 5-8 key evidence nodes, significantly enhancing decision transparency and credibility. Attached Figure Description
[0042] The present invention can be further illustrated by the non-limiting embodiments given in the accompanying drawings;
[0043] Figure 1 This is a system module connection diagram of an embodiment of a knowledge graph-driven multidimensional data acquisition and decision support system of the present invention;
[0044] Figure 2 This is a flowchart illustrating the steps of an embodiment of a knowledge graph-driven multidimensional data acquisition and decision support system according to the present invention.
[0045] The main component symbols are explained as follows: Data acquisition layer 101, knowledge graph layer 102, analysis engine layer 103, decision support layer 104, user interaction layer 105, feature extraction module 106, sentiment analysis submodule 107, multi-hypothesis reasoning unit 108, causal path visualization unit 109. Detailed Implementation
[0046] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments. The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0047] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.
[0048] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0049] like Figure 1As shown, the knowledge graph-driven multidimensional data acquisition and decision support system of the present invention includes a data acquisition layer 101, a knowledge graph layer 102, an analysis engine layer 103, a decision support layer 104, and a user interaction layer 105. Each layer realizes data interaction through a standardized interface.
[0050] The data acquisition layer is deployed on edge nodes and connected to the knowledge graph layer 102 through a Kafka message queue. It is configured with 23 protocol adaptation sub-modules and supports HTTP / HTTPS, MQTT and WebSocket access protocols.
[0051] The knowledge graph layer 104 adopts a Neo4j+Redis hybrid storage architecture, receives standardized data pushed by the data acquisition layer 101 through the REST API interface, and provides data services to the analysis engine layer 103.
[0052] The analysis engine layer 103 is deployed on a GPU computing node, calls knowledge graph layer data through the gRPC protocol, and is configured with 8×NVIDIA A100 GPUs to achieve parallel computing.
[0053] The decision support layer 104 uses the Django framework to build an inference engine and pushes decision suggestions to the user interaction layer 105 via the SOAP protocol.
[0054] The user interaction layer 105 is deployed on the CDN network, realizes real-time data visualization through the WebSocket protocol, and uses the React framework to build a responsive interface.
[0055] In practical applications of this embodiment, the data acquisition layer 101 includes a three-level filtering mechanism: filtering spam advertising content based on a rule base; identifying outliers using the isolated forest algorithm; and removing contradictory data using contextual semantic consistency verification.
[0056] It also includes a feature extraction module 106 connected to the data acquisition layer 101. The feature extraction module 106 customizes extraction schemes for different data types. Text data is represented by a fusion of TF-IDF and BERT embedding vectors. Time series data is extracted by wavelet transform to extract frequency domain features. Spatial data is converted into GeoHash encoding and coordinate normalization representation.
[0057] In the practical application of this embodiment, the analysis engine layer 103 includes a sentiment analysis submodule 107, which adopts a bidirectional encoder architecture, is pre-trained based on the BERT-base model and fine-tuned by the domain sentiment dictionary, introduces a context-aware attention mechanism, and supports explicit sentiment polarity judgment and implicit stance recognition.
[0058] In the practical application of this embodiment, the decision support layer 104 constructs a dynamic reasoning framework based on an improved Bayesian network.
[0059] In the practical application of this embodiment, the decision support layer 104 further includes a causal path visualization unit 109, which is used to calculate the influence of key nodes. The calculation formula is as follows:
[0060]
[0061] in, Let S(u) be the damping coefficient, S(u) be the influence score of node u, L(u) be the out-degree of node u, and N be the total number of nodes in the network.
[0062] In the practical application of this embodiment, the knowledge graph layer 102 adopts a partitioned storage strategy to control the query latency of hot entities to less than 8ms; the knowledge graph layer 102 also adopts a distributed graph database cluster architecture and configures a master-slave replication mechanism to achieve automatic fault switching, controlling the data synchronization latency to less than 100ms.
[0063] A knowledge graph-driven multidimensional data acquisition and decision support method, wherein the multi-granularity knowledge graph construction method of the knowledge graph layer 102 includes the following steps:
[0064] S1: Data access, receiving standardized feature vectors from data acquisition layer 101;
[0065] S2: Entity recognition, using an improved bidirectional LSTM-CRF model to perform entity recognition and disambiguation on multi-source data;
[0066] S3: Relation extraction, which extracts relationships between entities by combining syntactic dependency parsing and remote supervision techniques;
[0067] S4: Dynamic update, introducing a time-series decay factor to achieve minute-level updates of the knowledge graph and calculate entity weights.
[0068] In practical applications of this embodiment, the S4 dynamic update also includes a context-aware weight adjustment mechanism, which dynamically adjusts the knowledge weights using the health status parameter ct and the treatment stage parameter ht. The calculation formula is as follows:
[0069]
[0070] in, This is the entity importance coefficient. As a time-dependent decay factor, For health status parameters, The value range is 0-1. These are parameters for the diagnosis and treatment phase.
[0071] The working principle of this invention is as follows:
[0072] A multi-source heterogeneous data preprocessing architecture employs a pipelined processing approach, with each stage coordinating in parallel through a distributed task scheduling system. During the data access phase, a protocol parser cluster is configured to perform syntax validation on 15 file formats, including JSON, XML, and CSV, automatically triggering a format repair mechanism for abnormal formats. The format conversion module uses a combination of XSLT stylesheets and a custom parser, controlling the character error rate to within 3.7% when converting unstructured speech data (sampling rate 16kHz) to text. Image OCR recognition employs a multi-model fusion strategy, using a CNN+Transformer architecture to achieve approximately 98% character recognition accuracy. The noise filtering stage deploys a three-level filtering mechanism: filtering spam advertising content based on a rule base (95% coverage), identifying outliers using the Isolation Forest algorithm (89% detection rate), and removing contradictory data using contextual semantic consistency checks (false positive rate <5%). The feature extraction module 106 customizes extraction schemes for different data types: text data is represented by a fusion of TF-IDF and BERT embedding vectors, time series data is extracted by wavelet transform to extract frequency domain features, and spatial data is converted into GeoHash encoding and coordinate normalization representation, ultimately forming a 1024-dimensional standardized feature vector.
[0073] The knowledge graph dynamic update adopts an incremental processing framework, triggering update events through real-time data streams. The entity linking stage employs an improved bidirectional LSTM-CRF model to identify and disambiguate entities in newly added data, achieving an entity matching accuracy of 91.5% with an average processing time of 42ms. The relation extraction module integrates syntactic dependency parsing and remote supervision techniques. The weight adjustment mechanism introduces a time-decay factor, dynamically updating entity importance weights over time; the calculation formula is as follows:
[0074]
[0075] in, As the initial weights, The attenuation coefficient is... The value range is 0.012-0.058. Let be the correlation strength between entity e and event i at time t. Score the impact of the event.
[0076] The time-varying causal propagation network constructs a dynamic reasoning framework based on an improved Bayesian network. Network nodes include three types of elements: entities, relationships, and events, with edge weights representing the strength of causal associations. Causal path mining employs a combination of depth-first search and information entropy pruning algorithms, achieving second-level path discovery in a network with 10 nodes. Node influence calculation integrates PageRank and sentiment propagation models; the formula for the comprehensive influence score of node v at time t is:
[0077]
[0078] in, As the influence decay factor, 0.3 ≤ α ≤ 0.7, Let O(u,t) represent the causal strength from u to v at time t, O(u,t) be the out-degree of node u, and S(v,t) be the sentiment tendency value of node v.
[0079] The multi-hypothesis reasoning unit 108 uses the evidence theory framework to calculate hypothesis credibility; the multi-hypothesis reasoning algorithm flow is as follows:
[0080] Hypothesis generation: Based on event node triggering from a knowledge graph, a strategy combining template matching and generative models is adopted to automatically generate 3-5 possible evolution directions for public opinion events.
[0081] Credibility Assessment: An evidence theory framework is introduced, mapping multi-source data to a basic probability assignment function. The credibility of the hypothesis is calculated using Dempster's composition rule. The formula for calculating hypothesis credibility is:
[0082]
[0083] Among them, the conflict coefficient .
[0084] In summary, this embodiment achieves significant breakthroughs in key indicators such as data association efficiency, sentiment recognition accuracy, knowledge update speed, and system concurrency by setting up a data acquisition layer 101, a knowledge graph layer 102, an analysis engine layer 103, a decision support layer 104, and a user interaction layer 105. In terms of multi-source heterogeneous data processing, traditional systems using the ETL offline mode result in data association taking up to hours. This embodiment, through a distributed real-time processing architecture, shortens the association time to the second level, which can meet the needs of minute-level decision response.
[0085] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.
Claims
1. A knowledge graph-driven multidimensional data acquisition and decision support system, characterized in that: It includes a data acquisition layer (101), a knowledge graph layer (102), an analysis engine layer (103), a decision support layer (104), and a user interaction layer (105), with each layer realizing data interaction through standardized interfaces; The data acquisition layer is deployed on edge nodes and connected to the knowledge graph layer (102). It is configured with multiple protocol adaptation sub-modules and supports HTTP / HTTPS, MQTT and WebSocket access protocols. The knowledge graph layer (104) adopts a hybrid storage architecture, receives standardized data pushed by the data acquisition layer (101) through the REST API interface, and provides data services to the analysis engine layer (103); The analysis engine layer (103) is deployed on a GPU computing node, and calls the knowledge graph layer data through the gRPC protocol to configure the GPU to achieve parallel computing; The decision support layer (104) pushes decision suggestions to the user interaction layer (105) via the SOAP protocol; The user interaction layer (105) is deployed on the CDN network and uses the WebSocket protocol to achieve real-time data visualization.
2. The knowledge graph-driven multidimensional data acquisition and decision support system according to claim 1, characterized in that: The data acquisition layer (101) includes a three-level filtering mechanism: filtering ad spam content based on a rule base; identifying outliers through the isolated forest algorithm; and removing contradictory data using contextual semantic consistency verification. It also includes a feature extraction module (106) connected to the data acquisition layer (101). The feature extraction module (106) customizes extraction schemes for different data types. Text data is represented by a fusion of TF-IDF and BERT embedding vectors. Time series data is extracted by wavelet transform to extract frequency domain features. Spatial data is converted into GeoHash encoding and coordinate normalization representation.
3. The knowledge graph-driven multidimensional data acquisition and decision support system according to claim 1, characterized in that: The analysis engine layer (103) includes a sentiment analysis submodule (107), which adopts a bidirectional encoder architecture, is pre-trained based on the BERT-base model and fine-tuned by the domain sentiment dictionary, introduces a context-aware attention mechanism, and supports explicit sentiment polarity judgment and implicit stance recognition.
4. The knowledge graph-driven multidimensional data acquisition and decision support system according to claim 1, characterized in that: The decision support layer (104) constructs a dynamic reasoning framework based on an improved Bayesian network, and the node influence calculation formula is as follows: in, As an influence decay factor, Let O(u,t) represent the causal strength from u to v at time t, O(u,t) be the out-degree of node u, and S(v,t) be the sentiment tendency value of node v.
5. The knowledge graph-driven multidimensional data acquisition and decision support system according to claim 1, characterized in that: The decision support layer (104) further includes a multi-hypothesis reasoning unit (108), which calculates hypothesis credibility using an evidence theory framework. The hypothesis credibility calculation formula is as follows: Among them, the conflict coefficient .
6. The knowledge graph-driven multidimensional data acquisition and decision support system according to claim 1, characterized in that: The decision support layer (104) also includes a causal path visualization unit (109), which is used to calculate the influence of key nodes. The calculation formula is as follows: in, Let S(u) be the damping coefficient, S(u) be the influence score of node u, L(u) be the out-degree of node u, and N be the total number of nodes in the network.
7. The knowledge graph-driven multidimensional data acquisition and decision support system according to claim 1, characterized in that: The knowledge graph layer (102) adopts a partitioned storage strategy to control the query latency of hot entities to less than 8ms; the knowledge graph layer (102) also adopts a distributed graph database cluster architecture and is configured with a master-slave replication mechanism to achieve automatic fault switching, controlling the data synchronization latency to less than 100ms.
8. A knowledge graph-driven multidimensional data acquisition and decision support method, characterized in that, The multi-granularity knowledge graph construction method of the knowledge graph layer (102) includes the following steps: S1: Data access, receiving standardized feature vectors from the data acquisition layer (101); S2: Entity recognition, using an improved bidirectional LSTM-CRF model to perform entity recognition and disambiguation on multi-source data; S3: Relation extraction, which extracts relationships between entities by combining syntactic dependency parsing and remote supervision techniques; S4: Dynamic update, introducing a time-series decay factor to achieve minute-level updates of the knowledge graph and calculate entity weights. The entity weight adjustment formula is: in, As the initial weights, The attenuation coefficient is... The value range is 0.012-0.
058. Let be the correlation strength between entity e and event i at time t. Score the impact of the event.
9. A knowledge graph-driven multidimensional data acquisition and decision support system according to claim 8, characterized in that: The S4 dynamic update also includes a context-aware weight adjustment mechanism, which dynamically adjusts knowledge weights using the health status parameter ct and the treatment stage parameter ht. The calculation formula is as follows: in, This is the entity importance coefficient. As a time-dependent decay factor, For health status parameters, The value range is 0-1. These are parameters for the diagnosis and treatment phase.