CN111275079B - Crowd-sourced label presumption method and system based on graph neural network - Google Patents
Crowd-sourced label presumption method and system based on graph neural network Download PDFInfo
- Publication number
- CN111275079B CN111275079B CN202010034292.6A CN202010034292A CN111275079B CN 111275079 B CN111275079 B CN 111275079B CN 202010034292 A CN202010034292 A CN 202010034292A CN 111275079 B CN111275079 B CN 111275079B
- Authority
- CN
- China
- Prior art keywords
- task
- labeling
- label
- graph
- nodes
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a crowdsourcing label presumption method and a crowdsourcing label presumption system based on a graph neural network, wherein the crowdsourcing label presumption method comprises the following steps: (1) Carrying out data processing on the crowdsourcing labels to obtain initial characteristics of labeling personnel and tasks; (2) Constructing a labeling person-task abnormal graph and a labeling person isomorphic graph and a task isomorphic graph for task allocation conditions of labeling persons; (3) Inputting the labeling person-task abnormal graph, the labeling person isomorphic graph and the task isomorphic graph into a graph neural network to obtain embedded characteristics of task nodes; (4) And inputting the obtained embedded features of the task nodes into a prediction layer to obtain the probability that the task belongs to each label, and taking the label with the highest probability as the correct label of the task. The invention realizes crowdsourcing label speculation with high accuracy by using the graph neural network, is beneficial to generating a large amount of available machine learning training data, helps people train algorithm models and improves the competitiveness in the AI field.
Description
Technical Field
The invention relates to the field of label speculation in a crowdsourcing mode, in particular to a crowdsourcing label speculation method and system based on a graph neural network.
Background
Machine learning, particularly supervised learning, has been widely used in the fields of computer vision, natural language processing, and the like. Since supervised learning requires a large number of samples of known correct labels to train the model, the traditional approach is for the field specialist to review the samples and label them correctly, which is often expensive and time consuming and cannot meet the increasing demands on label data.
Crowdsourcing (crowds) is now one of the most important tools to obtain data tags due to its low cost and high efficiency features. People can conveniently acquire and utilize crowdsourcing resources by means of online platforms such as Amazon Mechanical Turk (AMT) and CrowdFlower. In these platforms, each task will be assigned to a different annotator, who returns tags for the task that are not necessarily correct (these tags are referred to as crowd-sourced tags, as distinguished from correct tags). Once labels of labeling people for a task are obtained, one intuitive strategy to obtain the correct labels of the task is a Majority Voting algorithm (Majority Voting), i.e., assuming that each labeling person has an equivalent number of votes, and the label with the highest number of votes obtained for the labeling person is considered as the correct label. However, crowdsourcing tags inevitably generate noise due to a variety of factors. For example, labeling personnel have varying degrees of expertise and reliability, while tasks have varying degrees of difficulty and confusion. Thus, simple strategies such as majority voting have difficulty in deriving reliable inferences, especially in cases where crowd-sourced tags are of low quality.
A series of findings of the existing crowdsourcing label speculation work indicate that modeling potential features of individual labels and tasks is critical. Based on the assumption that the key factors for deducing the correct labels are the ability of the annotators and the difficulty level of the task, researchers have proposed a number of probabilistic models that implement performance beyond the majority voting strategy. However, these models often require elaborate generation processes and inference algorithms for them and are also not suitable for large-scale datasets. Still other deep learning models attempt to learn both classifier models and label aggregation models, but these models require additional task features and rarely take into account bi-directional interactions between the annotators and the tasks.
Disclosure of Invention
The invention provides a crowdsourcing label speculation method based on a graph neural network, which utilizes the graph neural network to realize crowdsourcing label speculation with high accuracy, is favorable for generating a large amount of available machine learning training data, helps people train an algorithm model and improves the competitiveness in the AI field.
The specific technical scheme is as follows:
a crowdsourcing label presumption method based on a graph neural network comprises the following steps:
(1) Carrying out data processing on the crowdsourcing labels to obtain initial characteristics of labeling personnel and tasks;
(2) Constructing a labeling person-task abnormal graph and a labeling person isomorphic graph and a task isomorphic graph for task allocation conditions of labeling persons;
(3) Inputting the labeling person-task abnormal graph, the labeling person isomorphic graph and the task isomorphic graph into a graph neural network to obtain embedded characteristics of task nodes;
(4) And inputting the obtained embedded features of the task nodes into a prediction layer to obtain the probability that the task belongs to each label, and taking the label with the highest probability as the correct label of the task.
The crowdsourcing label speculation method uses crowdsourcing label data to construct a relation network for the labeling personnel and the tasks, and models the labeling personnel and the tasks, so that accurate labels of the tasks are accurately speculated by using implicit information of the labeling personnel and the tasks, implicit connections among the labeling personnel and implicit connections among the tasks.
The crowdsourcing label data is from labeling of tasks by labeling personnel, the number of labeling personnel is recorded as n, the number of tasks is recorded as m, and each labeling personnel provides labeling for tasks with unequal numbers (less than or equal to m).
In step (1), performing data processing on the crowdsourcing tag, including:
(1-1) for a labeling person, the initial feature dimension is a hyper-parameter, and each dimension is the probability that the label labeled by the labeling person is the same as the label obtained by majority voting;
(2-2) for a task, the initial feature dimension is a hyper-parameter, the initial feature dimension and the dimension of the initial feature of the labeling person are kept consistent, and each dimension is the probability that the labeling person labels the label of the task and the label obtained by majority voting are the same.
In the step (2):
the labeling person-task heterograph takes labeling person and task as nodes, the labeling person and the task are two types of nodes, and the edge between the labeling person node and the task node represents that the task is distributed to the labeling person;
the isomorphic graph of the labeling personnel takes the labeling personnel as nodes, each node is provided with a feature, an edge is arranged between two labeling personnel nodes of which the labeled task intersection is not an empty set, the edge is provided with the feature, and the similarity of the node attributes of the two labeling personnel is expressed;
the task isomorphic graph takes tasks as nodes, each node is provided with a feature, an edge is arranged between two task nodes of which the label person intersection is not an empty set, the edge is provided with the feature, and the similarity of the attributes of the two task nodes is expressed.
Further, the similarity of the two labeling personnel nodes is calculated as the probability that the labeled labels are the same in the intersection of the tasks labeled by the two labeling personnel; the similarity of the two task nodes is calculated as the probability that labels for the tasks are the same in the label personnel intersection for labeling the two tasks.
Preferably, the graphic neural network is formed by stacking three message passing layers; the first message transfer layer transfers information between nodes of different types and updates the hidden state of the nodes; the second message transfer layer transfers information between the same type of nodes and updates the hidden state of the nodes; the third messaging layer is identical to the first messaging layer.
Preferably, in step (3), the labeling person-task iso-graph is input to the first message passing layer, and the labeling person isomorphic graph and the task isomorphic graph are input to the second message passing layer.
Preferably, the prediction layer performs linear transformation on the input once and then performs a Softmax activation function to obtain the output; the dimensions of the output are the same as the total number of categories of tasks, and all dimensions add to 1.
Based on the same inventive concept, the invention also provides a crowdsourcing label presumption system based on the graph neural network, which comprises the following steps:
the data processing and mapping module is used for processing the original crowdsourcing label data and constructing a labeling person-task abnormal graph, a labeling person isomorphic graph and a task isomorphic graph;
the feature extraction module is used for extracting features of the labeling personnel and the tasks through the graph neural network based on the labeling personnel-task abnormal graph, the labeling personnel isomorphic graph and the task isomorphic graph;
and the label prediction module predicts the probability that the task belongs to each label based on the characteristics extracted by the task, and takes the label with the highest probability as the correct label of the task.
Preferably, the graphic neural network is formed by stacking three message passing layers; the first message transfer layer transfers information between nodes of different types and updates the hidden state of the nodes; the second message transfer layer transfers information between the same type of nodes and updates the hidden state of the nodes; the third messaging layer is identical to the first messaging layer.
Further, the labeling person-task abnormal graph is input to the first message transmission layer, and the labeling person isomorphic graph and the task isomorphic graph are input to the second message transmission layer.
Preferably, the label prediction module performs linear transformation on the input once and then obtains the output through a Softmax activation function; the dimensions of the output are the same as the total number of categories of tasks, and all dimensions add to 1.
Compared with the prior art, the invention has the beneficial effects that:
the invention uses the graph neural network technology in artificial intelligence to construct a graph model for labeling personnel and tasks in crowdsourcing label speculation problems, models the implicit information of the labeling personnel and the tasks, and uses three message transfer layers to explicitly use the implicit connection between the labeling personnel and the tasks and between the labeling personnel and between the tasks, thereby accurately speculating the correct labels of the tasks.
Drawings
FIG. 1 is a schematic diagram of a crowdsourcing tag speculation system according to an embodiment of the present invention;
FIG. 2 is a schematic workflow diagram of a crowdsourcing tag speculation system in accordance with an embodiment of the present invention;
FIG. 3 is a graph of a labeling person-task heterogram in accordance with an embodiment of the present invention;
FIG. 4 is a graph of isomorphism of a labeling person according to an embodiment of the present invention;
fig. 5 is a task isomorphic diagram of an embodiment of the present invention.
Detailed Description
The invention will be described in further detail below with reference to the drawings and examples, it being noted that the examples described below are intended to facilitate an understanding of the invention and are not intended to limit the invention in any way.
As shown in fig. 1, the crowdsourcing label speculation system according to the embodiment of the present invention includes a data processing and mapping module, a feature extraction module and a label prediction module, and the workflow thereof is as shown in fig. 2, and specifically includes the following steps: after the data is obtained, carrying out data preprocessing operation to obtain marked data, basic characteristics of tasks and data for graph modeling; then constructing a labeling person-task abnormal graph, a labeling person isomorphic graph and a task isomorphic graph by using graph modeling data, wherein the graph modeling data are respectively shown in fig. 3, 4 and 5; inputting the obtained labeling person-task abnormal graph, labeling person isomorphic graph and task isomorphic graph into a graph neural network, wherein the labeling person-task abnormal graph is input into a first message transmission layer in a graph neural network module, and the labeling person isomorphic graph and the task isomorphic graph are input into a second message transmission layer; obtaining embedded characteristics of nodes; and inputting the embedded characteristics of the task nodes into a prediction layer, and finally obtaining the correct labels of the tasks. The process is realized by a data processing and mapping module, a characteristic extraction module and a label prediction module step by step.
(1) The dataset web contains 2,665 tasks and 177 annotators, each annotating a task (not all tasks), the tasks having a total of 5 categories. A total of 15,567 tags were collected during this crowdsourcing process. We take 20% of the tasks and their crowdsourcing tags and correct tags as training data, and the remaining 80% of the tasks and their crowdsourcing tags and correct tags as test data.
(2) Firstly, carrying out data preprocessing operation on the original data as follows:
(2-1) for a labeling person, setting the initial feature dimension to be a super parameter, and setting the initial feature dimension to be 10, wherein each dimension is the probability that the label labeled by the labeling person is the same as the label obtained by majority voting;
(2-2) for a task, the initial feature dimension is a super parameter, and is set to 10 (consistent with the dimension of the initial feature of the labeling personnel), and each dimension is the probability that the labeling personnel label the task and the label obtained by majority voting are the same;
(3) The labeling personnel-task abnormal graph, the labeling personnel isomorphic graph and the task isomorphic graph are constructed by utilizing crowd-sourced label data, and the construction method is as follows:
(3-1) labeling person-task heterograms with all labeling persons and tasks as nodes, thus being an isomerism graph with two types of nodes, the initial characteristics of the nodes being set as the initial characteristics in step (2);
and (3-2) the labeling personnel isomorphic diagram takes all labeling personnel as nodes, and the edge attribute among the nodes is set as the similarity of the labeling personnel nodes. The similarity of the two labeling personnel nodes is calculated as the probability that labels labeled in the intersection set of the tasks labeled by the two labeling personnel are the same;
(3-3) the task isomorphic graph uses all tasks as nodes, and the edge attribute between the nodes is set as the similarity of the task nodes. The similarity of the two task nodes is calculated as the probability that labels marked on the tasks in the interaction of marking personnel marking the two tasks are the same;
(4) And inputting the obtained labeling person-task abnormal graph, labeling person isomorphic graph and task isomorphic graph into a graph neural network, wherein the graph neural network module comprises three message transmission layers as shown in fig. 1. The labeling personnel-task heterograms are input to a first message transmission layer in the graph neural network module and are used for transmitting messages among tasks labeled by the labeling personnel; the labeling person isomorphic diagram and the task isomorphic diagram are input to a second message transfer layer and used for transferring messages between the same class points, wherein node characteristics in the labeling person isomorphic diagram and the task isomorphic diagram are embedded characteristics output by the last message transfer layer; finally, inputting the generated embedded features into a third message transmission layer to generate final embedded features of the nodes;
(5) The embedded features of the task nodes are input into a prediction layer, and the task prediction classification probability is finally generated through linear transformation and Softmax activation functions.
The foregoing embodiments have described the technical solutions and advantages of the present invention in detail, and it should be understood that the foregoing embodiments are merely illustrative of the present invention and are not intended to limit the invention, and any modifications, additions, substitutions and the like that fall within the principles of the present invention should be included in the scope of the invention.
Claims (7)
1. The crowdsourcing label presumption method based on the graph neural network is characterized by comprising the following steps of:
(1) Carrying out data processing on the crowdsourcing labels to obtain initial characteristics of labeling personnel and tasks;
(2) Constructing a labeling person-task abnormal graph and a labeling person isomorphic graph and a task isomorphic graph for task allocation conditions of labeling persons;
(3) Inputting the labeling person-task abnormal graph, the labeling person isomorphic graph and the task isomorphic graph into a graph neural network to obtain embedded characteristics of task nodes;
the graphic neural network is formed by stacking three message transmission layers; the first message transfer layer transfers information between nodes of different types and updates the hidden state of the nodes; the second message transfer layer transfers information between the same type of nodes and updates the hidden state of the nodes; the third messaging layer is the same as the first messaging layer;
inputting the labeling person-task heterograms to a first message passing layer for passing messages between the labeling person and the labeled task; inputting the isomorphic diagrams of the labeling personnel and the isomorphic diagrams of the tasks into a second message transfer layer for transferring the messages among the similar nodes, wherein the node characteristics in the isomorphic diagrams of the labeling personnel and the isomorphic diagrams of the tasks are embedded characteristics output by the first message transfer layer; finally, the embedded feature generated by the second message transmission layer is input into the third message transmission layer to generate the embedded feature of the task node;
(4) And inputting the obtained embedded features of the task nodes into a prediction layer to obtain the probability that the task belongs to each label, and taking the label with the highest probability as the correct label of the task.
2. The method for crowdsourcing label speculation based on a graph neural network of claim 1, wherein in step (1), data processing is performed on crowdsourcing labels, comprising:
(1-1) for a labeling person, the initial feature dimension is a hyper-parameter, and each dimension is the probability that the label labeled by the labeling person is the same as the label obtained by majority voting;
(2-2) for a task, the initial feature dimension is a hyper-parameter, the initial feature dimension and the dimension of the initial feature of the labeling person are kept consistent, and each dimension is the probability that the labeling person labels the label of the task and the label obtained by majority voting are the same.
3. The method for crowdsourcing label speculation based on a graph neural network of claim 1, wherein in step (2):
the labeling person-task heterograph takes labeling person and task as nodes, the labeling person and the task are two types of nodes, and the edge between the labeling person node and the task node represents that the task is distributed to the labeling person;
the isomorphic graph of the labeling personnel takes the labeling personnel as nodes, each node is provided with a feature, an edge is arranged between two labeling personnel nodes of which the labeled task intersection is not an empty set, the edge is provided with the feature, and the similarity of the node attributes of the two labeling personnel is expressed;
the task isomorphic graph takes tasks as nodes, each node is provided with a feature, an edge is arranged between two task nodes of which the label person intersection is not an empty set, the edge is provided with the feature, and the similarity of the attributes of the two task nodes is expressed.
4. The crowdsourcing label speculation method based on a graph neural network of claim 3 wherein the similarity of node attributes of two annotators is calculated as the probability that the annotated labels are the same in the intersection of the task annotated by the two annotators; the similarity of the node attributes of the two tasks is calculated as the probability that labels for the tasks are the same in the label personnel intersection for labeling the two tasks.
5. The crowdsourcing label speculation method based on a graph neural network of claim 1 wherein the prediction layer performs a linear transformation on the input and then a Softmax activation function to obtain an output; the dimensions of the output are the same as the total number of categories of tasks, and all dimensions add to 1.
6. A crowdsourcing label speculation system based on a graph neural network, comprising:
the data processing and mapping module is used for processing the original crowdsourcing label data and constructing a labeling person-task abnormal graph, a labeling person isomorphic graph and a task isomorphic graph;
the feature extraction module is used for extracting features of the labeling personnel and the tasks through the graph neural network based on the labeling personnel-task abnormal graph, the labeling personnel isomorphic graph and the task isomorphic graph; the graphic neural network is formed by stacking three message transmission layers; the first message transfer layer transfers information between nodes of different types and updates the hidden state of the nodes; the second message transfer layer transfers information between the same type of nodes and updates the hidden state of the nodes; the third messaging layer is the same as the first messaging layer; inputting the labeling person-task heterograms to a first message passing layer for passing messages between the labeling person and the labeled task; inputting the isomorphic diagrams of the labeling personnel and the isomorphic diagrams of the tasks into a second message transfer layer for transferring the messages among the similar nodes, wherein the node characteristics in the isomorphic diagrams of the labeling personnel and the isomorphic diagrams of the tasks are embedded characteristics output by the first message transfer layer; finally, the embedded feature generated by the second message transmission layer is input into the third message transmission layer to generate the embedded feature of the task node;
and the label prediction module predicts the probability that the task belongs to each label based on the embedded characteristics of the task nodes, and takes the label with the highest probability as the correct label of the task.
7. The crowdsourcing label speculation system based on a neural network of claim 6 wherein the label prediction module performs a linear transformation on the input and then a Softmax activation function to obtain an output; the dimensions of the output are the same as the total number of categories of tasks, and all dimensions add to 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010034292.6A CN111275079B (en) | 2020-01-13 | 2020-01-13 | Crowd-sourced label presumption method and system based on graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010034292.6A CN111275079B (en) | 2020-01-13 | 2020-01-13 | Crowd-sourced label presumption method and system based on graph neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111275079A CN111275079A (en) | 2020-06-12 |
| CN111275079B true CN111275079B (en) | 2023-07-04 |
Family
ID=71002977
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010034292.6A Active CN111275079B (en) | 2020-01-13 | 2020-01-13 | Crowd-sourced label presumption method and system based on graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111275079B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112766337B (en) * | 2021-01-11 | 2024-01-12 | 中国科学院计算技术研究所 | Method and system for predicting correct tags for crowd-sourced data |
| CN113344387A (en) * | 2021-06-08 | 2021-09-03 | 浙江工商大学 | Crowdsourcing-based intelligent data processing method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684947A (en) * | 2018-12-11 | 2019-04-26 | 广州景骐科技有限公司 | Mark quality control method, device, computer equipment and storage medium |
| CN110097079A (en) * | 2019-03-29 | 2019-08-06 | 浙江工业大学 | A kind of privacy of user guard method based on classification boundaries |
| CN110209820A (en) * | 2019-06-05 | 2019-09-06 | 腾讯科技（深圳）有限公司 | User identifier detection method, device and storage medium |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9954942B2 (en) * | 2013-12-11 | 2018-04-24 | Entit Software Llc | Result aggregation |
| CN104573359B (en) * | 2014-12-31 | 2017-08-08 | 浙江大学 | A kind of mass-rent labeled data integration method of task based access control difficulty and mark person's ability |
| CN107357763B (en) * | 2017-06-12 | 2020-08-14 | 大连理工大学 | Crowdsourcing classification data quality control method based on self-walking learning |
| US11803883B2 (en) * | 2018-01-29 | 2023-10-31 | Nielsen Consumer Llc | Quality assurance for labeled training data |
| CN109284315B (en) * | 2018-08-24 | 2021-04-23 | 深圳莫比嗨客树莓派智能机器人有限公司 | Label data statistical inference method in crowdsourcing mode |
| CN110472054B (en) * | 2019-08-15 | 2023-05-23 | 北京爱数智慧科技有限公司 | Data processing method and device |
| CN110580499B (en) * | 2019-08-20 | 2022-05-24 | 北京邮电大学 | Deep learning target detection method and system based on crowdsourcing repeated labels |
-
2020
- 2020-01-13 CN CN202010034292.6A patent/CN111275079B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109684947A (en) * | 2018-12-11 | 2019-04-26 | 广州景骐科技有限公司 | Mark quality control method, device, computer equipment and storage medium |
| CN110097079A (en) * | 2019-03-29 | 2019-08-06 | 浙江工业大学 | A kind of privacy of user guard method based on classification boundaries |
| CN110209820A (en) * | 2019-06-05 | 2019-09-06 | 腾讯科技（深圳）有限公司 | User identifier detection method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111275079A (en) | 2020-06-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113822494B (en) | Risk prediction method, device, equipment and storage medium | |
| Li et al. | A perspective survey on deep transfer learning for fault diagnosis in industrial scenarios: Theories, applications and challenges | |
| US20230052865A1 (en) | Molecular graph representation learning method based on contrastive learning | |
| Tong | Active learning: theory and applications | |
| EP1903479B1 (en) | Method and system for data classification using a self-organizing map | |
| CN111275079B (en) | Crowd-sourced label presumption method and system based on graph neural network | |
| WO2024082453A1 (en) | Urban energy multi-source data fusion method based on deep learning and storage medium | |
| Nassehi et al. | Review of machine learning technologies and artificial intelligence in modern manufacturing systems | |
| CN113270199A (en) | Medical cross-modal multi-scale fusion class guidance hash method and system thereof | |
| CN116861924A (en) | Project risk early warning method and system based on artificial intelligence | |
| Shi et al. | Semi-supervised multi-label learning from crowds via deep sequential generative model | |
| CN112668633B (en) | Adaptive graph migration learning method based on fine granularity field | |
| Han et al. | DeepRouting: A deep neural network approach for ticket routing in expert network | |
| Ding et al. | Machinery cross domain degradation prognostics considering compound domain shifts | |
| Chakraborty et al. | Superensemble classifier for improving predictions in imbalanced datasets | |
| Haroon et al. | Application of machine learning in forensic science | |
| CN117236676A (en) | RPA process mining method and device based on multi-mode event extraction | |
| Jasim et al. | Analyzing Social Media Sentiment: Twitter as a Case Study | |
| Camci | Comparison of genetic and binary particle swarm optimization algorithms on system maintenance scheduling using prognostics information | |
| CN113342988A (en) | Method and system for constructing service knowledge graph to realize service combination optimization based on LDA cross-domain | |
| Yousefi | A Data-Driven Approach for Fault Classification of a Manufacturing Process | |
| Hu et al. | Data visualization analysis of knowledge graph application | |
| US20230419121A1 (en) | Systems and Methods for Programmatic Labeling of Training Data for Machine Learning Models via Clustering | |
| Babu et al. | Large dataset partitioning using ensemble partition-based clustering with majority voting technique | |
| CN108665056A (en) | A method of the Intelligent transfer robot based on NRL predicts task status |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |