CN115694877A - Space crowdsourcing task allocation method based on federal preference learning - Google Patents
Space crowdsourcing task allocation method based on federal preference learning Download PDFInfo
- Publication number
- CN115694877A CN115694877A CN202211046374.8A CN202211046374A CN115694877A CN 115694877 A CN115694877 A CN 115694877A CN 202211046374 A CN202211046374 A CN 202211046374A CN 115694877 A CN115694877 A CN 115694877A
- Authority
- CN
- China
- Prior art keywords
- task
- preference
- worker
- workers
- class
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 230000002146 bilateral effect Effects 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 239000013604 expression vector Substances 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 230000006399 behavior Effects 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims 1
- 230000008447 perception Effects 0.000 abstract description 2
- 238000011160 research Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Classifications
-
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a space crowdsourcing task allocation method based on federal preference learning. Then, the invention utilizes the federal learning framework to update the model parameters of the central server through the local model parameters of each platform center, thereby acquiring the global preference of workers. Finally, under the condition of considering the preference of workers, the task allocation is converted into the problem of maximum matching of bipartite graphs, the edge set of the established graphs is filtered by using a KM-based bilateral top-k intersection method, and the rest tasks are secondarily allocated by using a redistribution method, so that the total number of the allocated tasks is ensured while the matching algorithm is accelerated. The method can effectively obtain high task success rate and total number of the distributed tasks, and realize the task distribution of workers for task preference perception under the condition that the data of the multiple platforms of workers is not centralized, namely protect the privacy of the workers to a certain extent.
Description
Technical Field
The invention relates to the technical field of information, in particular to a space crowdsourcing task allocation method based on federal preference learning.
Background
Spatial crowdsourcing is ubiquitous in today's network world. The past decade has witnessed a tremendous advancement in space crowdsourcing that has enabled people to perform various location-based tasks with multimodal sensors. Social and ethical issues raised by spatial crowdsourcing are receiving increasing attention, including privacy and efficiency issues. Among them, privacy is one of the most common problems in spatial crowdsourcing. To enable efficient space crowdsourcing services, a worker or platform center (e.g., a company's delivery center) typically needs to disclose its original information (e.g., the worker's location and historical data). However, if the real data is utilized by a malicious third party, it is not only dangerous, but also makes people less willing to deliver the data to a spatial crowdsourcing platform, thereby resulting in low user participation and user loss.
Therefore, much research has begun to focus on privacy protection issues in spatial crowdsourcing, such as protecting worker's location information through encryption operations of worker location data perturbation, encryption, or distance calculation. However, they often only consider the protection of worker location information, ignoring worker preferences, and most often do not consider whether a worker is interested in a task's assignment method may result in the worker refusing to perform the task or completing the task with poor quality. Thus, such studies cannot guarantee a high quality of task allocation in reality and are not feasible.
In addition, there are many existing research methods that focus on the impact of various factors on task allocation, such as worker rejection, worker skill, platform profit, etc., to improve system feasibility. However, taking the preference factors of workers as an example, whether the task allocation problem of a single worker or the task allocation problem of a group is the case, these methods all acquire global worker preferences by establishing various complex preference models and performing centralized training on historical data and external data of all workers. However, in a real-life scene, a data island problem exists, and a mode of model centralized training is required, which cannot protect data of workers.
Thus, the existing methods all have some different drawbacks, namely that they cannot acquire global worker preferences with some protection of worker data. Therefore, the conventional research methods are no longer applicable.
Disclosure of Invention
The invention aims to provide a space crowdsourcing task allocation method based on federal preference learning.
In order to achieve the purpose, the invention is implemented according to the following technical scheme:
the method comprises a federal preference learning stage and a task allocation stage, wherein the federal preference learning stage is used for carrying out model training on a local preference model by using federal learning, so that a final preference model of a server center end is obtained, and further, the preference of global workers is obtained; and in the task allocation stage, the task allocation process is accelerated by a KM-based bilateral top-k intersection method and a task reallocation method, and higher task allocation quantity and completion rate are ensured.
Further, the local preference model is constructed by a context encoder module, an intra-class encoder module and a cooperation module, the context encoder module acquires context information containing task sequences and task category information, and the model predicts the category of the next task to be used for determining which categories of worker preferences should be utilized; the intra-class encoder module will further predict the preference of workers in each class for the next task location; the collaboration module obtains preferences of similar workers to assist in predicting the preferences of a target worker.
The beneficial effects of the invention are:
the invention relates to a space crowdsourcing task allocation method based on federal preference learning. Preference modeling is first performed on local data stored at each platform center. The local preference model obtains the preference of workers to task positions and task categories by utilizing a context encoder, an intra-class encoder and a cooperation module. Then, the invention utilizes the federal learning framework to update the model parameters of the central server through the local model parameters of each platform center, thereby acquiring the global preference of workers. Finally, under the condition of considering the preference of workers, the task allocation is converted into the problem of maximum matching of bipartite graphs, the edge set of the established graphs is filtered by using a KM-based bilateral top-k intersection method, and the rest tasks are secondarily allocated by using a redistribution method, so that the total number of the allocated tasks is ensured while the matching algorithm is accelerated. Therefore, the invention can effectively obtain high task success rate and total number of distributed tasks, and realize the task distribution of workers sensing the task preference under the condition of non-centralized data of multiple platforms, namely protect the privacy of the workers to a certain extent.
Drawings
FIG. 1 is a general architecture diagram of spatial crowdsourcing task allocation for Federal preference learning based on the present invention;
FIG. 2 is a platform centric preference model in the present invention.
Detailed Description
The invention will be further described with reference to the drawings and specific embodiments, which are illustrative of the invention and are not to be construed as limiting the invention.
As shown in fig. 1 and 2: the whole network architecture of the invention is divided into two stages: a federal preference learning phase and a task allocation phase.
In the federal preference learning stage, the local preference model needs to be subjected to model training by using federal learning, so that a final preference model of the server center end is obtained, and the preference of global workers is obtained.
In the construction of the local preference model, the model mainly comprises three modules, namely a context encoder module, an intra-class encoder module and a cooperation module.
First, the context encoder is used to obtain context information containing task sequence and task category information. The model predicts the categories of the next task to determine which worker preferences within which categories should be utilized.
To obtain category context information, the module uses a top-k gated network. For a top-k gating network, a recent task category sequence of a worker is used as input, and an embedding vector X is obtained by combining the category embedding layer and the position embedding layer to be respectively mapped z And relative position vector P z . Then, the two are added and input into the attention network to obtain the corresponding implicit expression vector
Then, the top-k gated network generates a probability distribution for all task classes:
wherein ,
indicating the probability that the next task class is class j,
is a category embedded representation of category j. Considering the uncertainty of prediction, the top-k gating network selects the top-k task categories with the highest probability as prediction candidates of the next task category of the worker:
wherein the first k task categories c are selected j E.g., C, and C represents the set of all task categories.
To capture the task sequence context information hidden in the recent task records, the module also employs a recent activity encoder. Similarly, recent behavior encodingThe device takes a task sequence of a worker in the near term as input, and combines the task site embedding layer and the position embedding layer to respectively map to obtain an embedding vector X r And relative position vector P r . Then, the context information expression vector of the task sequence presumed by adding the two is obtained from the attention network
Then, k predicted task categories are obtained
The intra-class encoder module will then further predict the worker's preference for the next task location within each class. Taking class c as an example, the sub-sequences of task sites under the corresponding class are mapped into a vector X through the task site embedding layer c Obtaining a relative position vector P by using the relative position information and the position embedding layer in the subsequence c In conjunction with the input to the self-attention network, a representation of the worker's preference in category c may be obtained
Then, in order to alleviate sparseness of the interaction data, the utilization cooperation module can acquire the preference of similar workers to assist in predicting the preference of the target worker. Wherein the in-class preference of the target worker w is given
The similarity of worker i to worker w is calculated as follows:
and then selecting the vector representations of top-f similar workers for weighted summation to serve as the final vector representation of the in-class preference of the similar workers
Wherein the selection range of top-F similar workers is F workers who have recently moved in the category.
Based on the three modules, the obtained task sequence context information, task category information and in-category preference information can be used for obtaining the preference of the target worker on the task location and the task category. Task class c for target worker j The prediction preference value of (a) is calculated as follows:
wherein ,
category information representing a recent task sequence of a worker,
is a category embedded representation of category j.
To predict the target worker versus task location l i The vector representation of the three kinds of information obtained before is connected and the final vector h representing the target worker to the category j is obtained through a full-connection network layer j ：
wherein ,
it is shown that the connection operation is performed,
context information indicative of recent task sequences for the worker,
in-class preference information indicating a worker,
represents the in-class preference of similar workers, and j =1,2.
The task location/of the worker under the category j can be obtained by calculating i Preference (c):
wherein ,Eout An embedded representation representing all task places.
When all top-k categories are considered, then the target worker is on task site l i The preference of (c) is calculated as follows:
after the local preference model of each platform center is constructed, the local preference model is subjected to model training by using federal learning, and a loss function l (θ) in the federal learning can be expressed as follows:
where m represents the number of platform centers selected, L k (θ) represents a loss function for the selected platform center k, which is calculated as follows:
where θ represents a set of parameters of the preference model, L loc Represents the loss of the task site prediction, L cate Represents the loss of the task class prediction, and the over parameter lambda is the weight for controlling the two losses.
To speed up the calculation of predicted loss at a task site, for each real task site, N is randomly assumed based on the popularity of the task site s The individual task sites serve as negative examples. L is loc The calculation formula of (c) is as follows:
where δ (-) is an indicator function, l N+1 A real task location is represented and,
the task location predicted by the model is represented.
Similarly, a loss function L may be calculated cate ：
During a particular federal training procedure, the following operations will be repeated for each round t: first, toAll platform centers are sampled to obtain a platform center set P participating in training t (ii) a Then, the server center sends the current model parameter theta t-1 Transmitting to each selected platform center; then, for each platform center not participating in training, the model parameters are kept constant, while each platform center pc participating in training k Then model local training is performed:
after the local training is finished, each platform center participating in the training updates the parameters locally
And transmitting to the central server. Then, the central server collects the updated parameters of each center, updates the model parameters of the server center, and enables the updating conditions to meet the following formula:
in a task allocation stage, a KM-based bilateral top-k intersection method and a task reallocation method are provided to accelerate a task allocation process and ensure higher task allocation quantity and completion rate.
Firstly, the algorithm converts the task allocation problem into a bipartite graph maximum weight matching problem by establishing a bipartite graph of workers and tasks. The point set in the created bipartite graph is divided into V W and VS Two sets of workers w each i Mapping to a vertex
Each space task s j Mapping to a vertex
The edge sets are added according to the space-time limit and the top-k limit:
(1) Space-time limitation: w is a i ∈AW(s j) and sj ∈RT(w i )
(2) top-k restriction:
and 
Wherein AW(s) j ) And RT (w) i ) Respectively represent pair tasks s j Set of available workers meeting space-time constraints, for worker w i A set of reachable tasks that meet spatiotemporal constraints. For any W ∈ AW (S), S ∈ S or S ∈ RT (W), W ∈ W should satisfy the following two conditions:
(1)d(w.l,s.l)≤w.r
(2)t now +d(w.l,s.l)/w.speed≤s.e
the above two conditions respectively represent: task s needs to be within reach of worker w; workers need to arrive at the execution site before the task is out of date.
From the worker
Connecting to space tasks
The weight of the edge is given by the worker w i For task s j Is measured by the preference of (1) and recorded as
The specific calculation is as follows:
wherein ,
indicating worker w i The preference for all the categories of the task,
representing a task s j The embedded vector of the corresponding category.
To better implement the task assignment algorithm, the algorithm limits the recursion depth to k in the function that recursively finds the task that the worker can match. At the same time, the function computes the difference between the weight of the edge associated with the two vertices and the sum of the expected values for the worker and the task. If the difference is equal to 0, the task may be assigned to the worker. Where the expected values for workers and tasks are the maximum weight values in the edges associated therewith. When the worker fails to match any task, the expectation of the worker and the task involved in the last matching is adjusted, and the competition relationship among the workers is changed, so that the aim of distributing more workers is fulfilled. Furthermore, since the worker task bipartite graph may not have the reality of a perfect match, for workers whose expected value is less than 0, we will stop matching tasks for that worker.
The top-k constraint significantly filters many low-weight edges, resulting in a large loss of task assignment number when creating edge sets for bipartite graphs. Therefore, the present algorithm also incorporates a method of task reallocation to compensate for this loss, i.e. when there are tasks that are not allocated and there are still workers available, they are allocated to the worker that has the greatest preference for it.
The invention has the advantages that: under the condition that each platform center does not directly transmit data, effective modeling is carried out on task preference of workers through federal preference learning, and a KM-based bilateral top-k intersection method and a task reallocation method are combined, so that the task allocation process is greatly accelerated, and higher task allocation quantity and completion rate are ensured.
First, the present invention performs preference modeling on local data stored in each platform center. In the local preference model, context information in a worker-task sequence, the preference of a worker for a task location and the task location preference of a similar worker can be obtained by utilizing a context encoder, an intra-class encoder and a coordination module respectively. Then, the worker's preference for task location and task category can be obtained through a fully connected network in combination with the three obtained information. Then, in order to protect the data privacy of each platform center, the invention utilizes a federal learning framework to update the model parameters of the central server by transmitting the local model parameters of each platform center, thereby obtaining the global worker preference. And finally, under the condition of considering the preference of workers, the task allocation is converted into the problem of maximum matching of bipartite graphs, and the edge set of the created graph is filtered by using a KM-based bilateral top-k intersection method. Compared with the original KM method and the unilateral top-k method, the method ensures fast and high-quality task allocation. Meanwhile, the invention uses the redistribution method to carry out secondary distribution on the rest tasks, thereby ensuring higher total number of distributed tasks. Therefore, the invention can effectively obtain high task success rate and total number of the distributed tasks, and realize the task distribution of workers for task preference perception under the condition that the data of the multiple platforms of workers is not centralized, namely, the privacy of the workers is protected to a certain extent.
The technical solution of the present invention is not limited to the above-mentioned specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.
Claims (8)
1. A space crowdsourcing task allocation method based on federal preference learning is characterized in that: the method comprises a federal preference learning stage and a task allocation stage, wherein the federal preference learning stage is used for carrying out model training on a local preference model by using federal learning, so that a final preference model of a server center end is obtained, and further the preference of global workers is obtained; and in the task allocation stage, the task allocation process is accelerated by a KM-based bilateral top-k intersection method and a task reallocation method, and higher task allocation quantity and completion rate are ensured.
2. The federal preference learning-based spatial crowd-sourced task allocation method of claim 1, wherein: the local preference model is constructed by a context encoder module, an intra-class encoder module and a cooperation module, wherein the context encoder module acquires context information containing a task sequence and task class information, and the model predicts the class of the next task to be used for determining the worker preference in which classes should be utilized; the intra-class encoder module will further predict the preference of workers in each class for the next task location; the collaboration module obtains preferences of similar workers to assist in predicting the preferences of a target worker.
3. The federal preference learning-based spatial crowd-sourced task allocation method of claim 2, wherein: the context encoder module uses a top-k gating network; taking recent task category sequences of workers as input, and respectively mapping by combining a category embedding layer and a position embedding layer to obtain an embedding vector X z And a relative position vector P z (ii) a Then, the two are added and input into the attention network to obtain a corresponding implicit expression vector h z ：
h z ＝SelfAttention(X z +P z )
Then, the top-k gated network generates a probability distribution for all task classes:
wherein ,
indicating the probability that the next task class is class j,
is a category embedding table of the kind jShown in the specification; considering the uncertainty of prediction, the top-k gating network selects the top-k task categories with the highest probability as prediction candidates of the next task category of the worker:
wherein the first k selected task categories c j E.g., C, and C represents the set of all task categories.
4. The method of claim 3, wherein the spatial crowdsourcing task allocation based on federated preference learning is characterized in that: the context encoder module also adopts a recent behavior encoder, the recent behavior encoder takes a recent task sequence of a worker as input, and an embedded vector X is obtained by combining the task place embedded layer and the position embedded layer and respectively mapping r And relative position vector P r (ii) a Then, the context information of the task sequence is added and input to the expression vector h of the context information of the task sequence which is obtained by the inference from the attention network r ：
h r ＝SelfAttention(X r +P r )
Then, k predicted task categories are obtained
The intra-class encoder module will then further predict the worker's preference for the next task location within each class.
5. The method of claim 4, wherein the spatial crowdsourcing task allocation based on federated preference learning is characterized in that: in the intra-class encoder module, taking class c as an example, the task place subsequences under the corresponding classes are mapped into vector X through the task place embedding layer c Obtaining a relative position vector P by using the relative position information and the position embedding layer in the subsequence c In conjunction with the input to the self-attention network, a preference representation h for workers in category c may be obtained c ：
h c ＝SelfAttention(X c +P c )
Then, in order to alleviate sparseness of the interaction data, the collaborative module can be used for acquiring the preference of similar workers to assist in predicting the preference of the target worker.
6. The method of claim 5, wherein the spatial crowdsourcing task allocation based on federated preference learning is characterized in that: in the collaboration module, the in-class preferences of the target worker w are given
The similarity of worker i to worker w is calculated as follows:
then, selecting top-f vector representations of similar workers to carry out weighted summation to serve as final vector representation h of in-class preference of the similar workers f ：
Wherein the selection range of top-F similar workers is F workers who have recently moved in the category.
7. The method of claim 6, wherein the spatial crowdsourcing task allocation based on federated preference learning is characterized in that: the target worker to task class c j The prediction preference value of (c) is calculated as follows:
wherein ,hz Category information representing a recent sequence of tasks for the worker,
is a category embedded representation of category j;
to predict benchmarking worker versus task location l i The vector representation of the three kinds of information obtained before is connected and the final vector h representing the target worker to the category j is obtained through a full-connection network layer j ：
wherein ,
denotes a connection operation, h r Context information indicative of recent task sequences of the worker,
in-class preference information indicating a worker,
represents in-class preferences of similar workers, and j =1,2.. K;
the task location/of the worker under the category j can be obtained by calculating i Preference (c):
wherein ,Eout An embedded representation representing all task places;
when all top-k categories are considered, then the target worker is to task site l i The preference of (c) is calculated as follows:
after the local preference model of each platform center is constructed, the local preference model is subjected to model training by using federal learning, and a loss function l (theta) in the federal learning can be expressed as follows:
where m represents the number of selected platform centers, L k (θ) represents a loss function for the selected platform center k, which is calculated as follows:
where θ represents a set of parameters of the preference model, L loc Represents the loss of the task site prediction, L cate The loss of the task class prediction is represented, and the over parameter lambda is used for controlling the weight of the two losses;
to speed up the calculation of predicted loss at a task site, for each real task site, N is randomly assumed based on the popularity of the task site s The individual task sites are used as negative samples; l is loc The calculation formula of (c) is as follows:
where δ (·) is an indicator function, l N+1 A real task location is represented and,
representing the task location predicted by the model;
similarly, a loss function L may be calculated cate ：
During a particular federal training procedure, the following operations will be repeated for each round t: firstly, sampling all platform centers to obtain a platform center set P participating in training t (ii) a Then, the server center sends the current model parameter theta t-1 Transmitting to each selected platform center; then, for each platform center not participating in the training, the model parameters thereof remain unchanged, while each platform center pc participating in the training k Then model local training is performed:
after the local training is finished, each platform center participating in the training updates the parameters locally
Transmitting to a central server; then, the central server collects the updated parameters of each center, updates the model parameters of the server center, and enables the updating conditions to meet the following formula:
8. the method of claim 1, wherein the spatial crowdsourcing task allocation based on federated preference learning is characterized in that: in the task allocation stage, a task allocation problem is converted into a bipartite graph maximum weight matching problem by establishing a bipartite graph of workers and tasks; the point set in the created bipartite graph is divided into V W and VS Two sets of workers w each i Mapping to a vertex
Each space task s j Mapping to a vertex
The edge sets are then added according to the spatio-temporal constraints and top-k constraints:
(1) Space-time limitation: w is a i ∈AW(s j) and sj ∈RT(w i )
(2) top-k restriction:
and 
Wherein AW(s) j ) And RT (w) i ) Respectively representing the pair tasks s j Set of available workers, pair of workers w, meeting space-time constraints i A reachable task set that meets spatiotemporal constraints; for any W ∈ AW (S), S ∈ S or S ∈ RT (W), W ∈ W should satisfy the following two conditions:
(1)d(w.l,s.l)≤w.r
(2)t now +d(w.l,s.l)/w.speed≤s.e
the above two conditions respectively represent: task s needs to be within reach of worker w; workers need to arrive at the execution site before the task expires;
from the worker
Connecting to space tasks
The weight of the edge is given by the worker w i For task s j Is measured by the preference of (1) and recorded as
The specific calculation is as follows:
wherein ,
indicating worker w i The preference for all the categories of the task,
representing a task s j The embedded vector of the corresponding category.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211046374.8A CN115694877B (en) | 2022-08-30 | 2022-08-30 | Space crowdsourcing task allocation method based on federal preference learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211046374.8A CN115694877B (en) | 2022-08-30 | 2022-08-30 | Space crowdsourcing task allocation method based on federal preference learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115694877A true CN115694877A (en) | 2023-02-03 |
| CN115694877B CN115694877B (en) | 2023-08-15 |
Family
ID=85061000
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211046374.8A Active CN115694877B (en) | 2022-08-30 | 2022-08-30 | Space crowdsourcing task allocation method based on federal preference learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115694877B (en) |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019006966A1 (en) * | 2017-07-03 | 2019-01-10 | 深圳大学 | Task allocation system model of privacy protected spatial crowdsourcing, and implementation method |
| CN110400128A (en) * | 2019-07-29 | 2019-11-01 | 电子科技大学 | A kind of space crowdsourcing method for allocating tasks based on the perception of worker's preference |
| CN111291973A (en) * | 2020-01-17 | 2020-06-16 | 麦科思(苏州)数据科技有限公司 | Space crowdsourcing task allocation method based on alliance |
| CN111311115A (en) * | 2020-03-12 | 2020-06-19 | 电子科技大学 | Group task allocation method based on space crowdsourcing social influence preference |
| CN111866954A (en) * | 2020-07-21 | 2020-10-30 | 重庆邮电大学 | User selection and resource allocation method based on federal learning |
| CN113094758A (en) * | 2021-06-08 | 2021-07-09 | 华中科技大学 | Gradient disturbance-based federated learning data privacy protection method and system |
| CN113255966A (en) * | 2021-04-28 | 2021-08-13 | 南京邮电大学 | Crowdsourcing task matching method considering preference and independent places |
| WO2021189906A1 (en) * | 2020-10-20 | 2021-09-30 | 平安科技（深圳）有限公司 | Target detection method and apparatus based on federated learning, and device and storage medium |
| CN113887748A (en) * | 2021-12-07 | 2022-01-04 | 浙江师范大学 | Online federal learning task allocation method and device, and federal learning method and system |
| WO2022020968A1 (en) * | 2020-07-30 | 2022-02-03 | Interaxon Inc. | Learning representations of eeg signals with self-supervised learning |
| CN114048361A (en) * | 2021-11-11 | 2022-02-15 | 东南大学 | Crowdsourcing software developer recommendation method based on deep learning |
| CN114372680A (en) * | 2021-12-23 | 2022-04-19 | 电子科技大学(深圳)高等研究院 | Spatial crowdsourcing task allocation method based on worker loss prediction |
| CN114493106A (en) * | 2021-12-23 | 2022-05-13 | 电子科技大学(深圳)高等研究院 | Space crowdsourcing task allocation method based on geographical division |
| CN114580662A (en) * | 2022-02-28 | 2022-06-03 | 浙江大学 | Federal learning method and system based on anchor point aggregation |
| CN114579892A (en) * | 2022-05-06 | 2022-06-03 | 南京航空航天大学 | User remote access position prediction method based on cross-city interest point matching |
| CN114817663A (en) * | 2022-05-05 | 2022-07-29 | 杭州电子科技大学 | Service modeling and recommendation method based on class perception graph neural network |
-
2022
- 2022-08-30 CN CN202211046374.8A patent/CN115694877B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019006966A1 (en) * | 2017-07-03 | 2019-01-10 | 深圳大学 | Task allocation system model of privacy protected spatial crowdsourcing, and implementation method |
| CN110400128A (en) * | 2019-07-29 | 2019-11-01 | 电子科技大学 | A kind of space crowdsourcing method for allocating tasks based on the perception of worker's preference |
| CN111291973A (en) * | 2020-01-17 | 2020-06-16 | 麦科思(苏州)数据科技有限公司 | Space crowdsourcing task allocation method based on alliance |
| CN111311115A (en) * | 2020-03-12 | 2020-06-19 | 电子科技大学 | Group task allocation method based on space crowdsourcing social influence preference |
| CN111866954A (en) * | 2020-07-21 | 2020-10-30 | 重庆邮电大学 | User selection and resource allocation method based on federal learning |
| WO2022020968A1 (en) * | 2020-07-30 | 2022-02-03 | Interaxon Inc. | Learning representations of eeg signals with self-supervised learning |
| WO2021189906A1 (en) * | 2020-10-20 | 2021-09-30 | 平安科技（深圳）有限公司 | Target detection method and apparatus based on federated learning, and device and storage medium |
| CN113255966A (en) * | 2021-04-28 | 2021-08-13 | 南京邮电大学 | Crowdsourcing task matching method considering preference and independent places |
| CN113094758A (en) * | 2021-06-08 | 2021-07-09 | 华中科技大学 | Gradient disturbance-based federated learning data privacy protection method and system |
| CN114048361A (en) * | 2021-11-11 | 2022-02-15 | 东南大学 | Crowdsourcing software developer recommendation method based on deep learning |
| CN113887748A (en) * | 2021-12-07 | 2022-01-04 | 浙江师范大学 | Online federal learning task allocation method and device, and federal learning method and system |
| CN114372680A (en) * | 2021-12-23 | 2022-04-19 | 电子科技大学(深圳)高等研究院 | Spatial crowdsourcing task allocation method based on worker loss prediction |
| CN114493106A (en) * | 2021-12-23 | 2022-05-13 | 电子科技大学(深圳)高等研究院 | Space crowdsourcing task allocation method based on geographical division |
| CN114580662A (en) * | 2022-02-28 | 2022-06-03 | 浙江大学 | Federal learning method and system based on anchor point aggregation |
| CN114817663A (en) * | 2022-05-05 | 2022-07-29 | 杭州电子科技大学 | Service modeling and recommendation method based on class perception graph neural network |
| CN114579892A (en) * | 2022-05-06 | 2022-06-03 | 南京航空航天大学 | User remote access position prediction method based on cross-city interest point matching |
Non-Patent Citations (9)
| Title |
|---|
| ESAM ALOUFI等: "An Efficient Approach for Task Assignment in Spatial Crowdsourcing", 《2020 IEEE INTERNATIONAL IOT，ELECTRONICS AND MECHATRONICS CONFERENCE（IEMTRONICS）》 * |
| WANGZE NI等: "Task Allocation in Dependency-aware Spatial Crowdsourcing", 《2020 IEEE 36TH INTERNATIONAL CONFERENCE ON DATA ENGINEERING（ICDE）》 * |
| YUNCHUAN LI，YAN ZHAO，KAI ZHENG: "Preference-aware Group Task Assignment in Spatial Crowdsourcing：A Mutual Information-based Approach", 《2021 IEEE INTERNATIONAL CONFERENCE ON DATA MINING（ICDM）》 * |
| 刘雪娇，王慧敏等: "具有隐私保护的车联网空间众包任务分配方法", 《浙江大学学报》, no. 07 * |
| 柴成亮;李国良;赵天宇;骆昱宇;于明鹤;: "众包数据库综述", 计算机学报, no. 05 * |
| 胡颖;王莹洁;童向荣;: "基于众包工人移动轨迹的任务推荐模型", 计算机科学, no. 10 * |
| 谢珍真: "基于深度学习的众包计算资源分配与群体行为异构性研究", 《中国博士学位论文全文数据库 信息科技辑》, no. 01 * |
| 赵祺雯;彭瑞;袁平;: "基于注意力机制的众包任务推荐算法", 现代计算机, no. 03 * |
| 赵鹏程，高尚，于洪梅: "基于多智能体深度强化学习的空间众包任务分配", 《吉林大学学报》, no. 02 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115694877B (en) | 2023-08-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11393341B2 (en) | Joint order dispatching and fleet management for online ride-sharing platforms | |
| CN108133330B (en) | Social crowdsourcing task allocation method and system | |
| CN106875066B (en) | Vehicle travel behavior prediction method, device, server and storage medium | |
| Shi et al. | ParkCrowd: Reliable crowdsensing for aggregation and dissemination of parking space information | |
| Fan et al. | Joint scheduling and incentive mechanism for spatio-temporal vehicular crowd sensing | |
| CN111353092B (en) | Service pushing method, device, server and readable storage medium | |
| WO2016035091A1 (en) | Dynamic forecasting for forward reservation of cab | |
| CN113901911B (en) | Image recognition method, image recognition device, model training method, model training device, electronic equipment and storage medium | |
| CN116823535B (en) | Journey planning and intelligent navigation system based on multi-mode large model | |
| CN112668880A (en) | Work order scheduling method and device for distribution network grid, computer equipment and storage medium | |
| CN107590548A (en) | Power consumer work dispatching method and storage medium | |
| CN111400747B (en) | Measurement method based on track privacy protection | |
| CN112035548A (en) | Identification model acquisition method, identification method, device, equipment and medium | |
| CN105786941A (en) | Information mining method and device | |
| CN114372680A (en) | Spatial crowdsourcing task allocation method based on worker loss prediction | |
| CN105825333A (en) | Crowdsourcing service system based on anonymous places of cloud platform and task distribution method | |
| US20210406740A1 (en) | Method and system for estimating relocation costs | |
| CN112052572A (en) | Digital twin industrial simulation system based on WLAN (Wireless local area network) position perception | |
| CN115694877A (en) | Space crowdsourcing task allocation method based on federal preference learning | |
| CN111353093B (en) | Problem recommendation method, device, server and readable storage medium | |
| CN116049678A (en) | Feature contribution degree evaluation method, device, electronic equipment and storage medium | |
| CN112258009B (en) | Intelligent government affair request processing method | |
| CN114969576A (en) | Method for training journey planning model and journey planning method | |
| WO2021229625A1 (en) | Learning device, learning method, and learning program | |
| CN111832765B (en) | Scheduling method, scheduling device, electronic equipment and readable storage medium |
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 |