Crowdsourcing task allocation method based on multi-element participant preferences
By constructing a multi-dimensional preference model and a Logit model, the problem of existing technologies failing to effectively consider users' multi-dimensional preferences and subjective intentions is solved, achieving more accurate and efficient group intelligence perception task allocation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2024-08-01
- Publication Date
- 2026-07-21
AI Technical Summary
Existing crowd-sensing task allocation methods fail to effectively consider users' multi-dimensional preferences and subjective intentions, resulting in inaccurate task allocation.
By constructing content preference models, time preference models, and location preference models based on text and image data, and combining them with the Logit model, multi-dimensional preferences are integrated to predict the match between participants and tasks, and to assign tasks.
It improves the accuracy and efficiency of task allocation, ensuring that tasks are assigned to the most suitable participants.
Smart Images
Figure CN119027955B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of crowd sensing technology, and in particular to a crowd sensing task allocation method based on the preferences of multiple participants. Background Technology
[0002] With the rapid development of mobile devices and sensor technologies, we are entering a data-driven era. Mobile Crowd Sensing (MCS), as an emerging sensing mode, is gradually becoming one of the key technologies for solving the problem of large-scale data collection and processing. MCS achieves real-time environmental perception and dynamic data collection through the participation of widely distributed mobile devices and users, providing strong data support for smart cities, environmental monitoring, traffic management, and other fields. However, the successful implementation of MCS faces many challenges, among which efficient task allocation is one of the key issues. How to rationally allocate tasks among numerous mobile devices and users to maximize the sensing effect has become an urgent problem to be solved.
[0003] Existing research on task allocation has proposed many algorithms, which can be broadly categorized into two types: platform-centric task allocation and user-centric task allocation. Platform-centric task allocation employs methods such as optimization-based, auction-based, prediction-based, and social network-based approaches. User-centric task allocation primarily uses methods based on participant preferences. However, these studies still have some shortcomings: 1. In platform-centric task allocation, existing research generally assumes that users will accept tasks assigned by the platform. In reality, users' subjective willingness has a significant impact on task acceptance. Furthermore, these studies generally highly simplify tasks, using only a few attributes (such as time, location, and reward) to define a task. However, with the increasing complexity of crowd-sensing tasks, some complex tasks are difficult to describe using only a few attributes. 2. In user-centered task allocation, existing research generally uses text data when utilizing historical task data. In practice, with technological advancements, users often generate multi-modal data, such as text, images, and sound, when completing crowdsourcing tasks. Furthermore, in many cases, the data from different modalities complement each other; only by integrating all data can the user's intent be accurately captured. Simultaneously, existing research mostly considers only a single dimension of preference when analyzing user preferences, while user preferences may be multi-dimensional. Summary of the Invention
[0004] The purpose of this application is to provide a crowd-sensing task allocation method based on the preferences of multiple participants. This method aims to address the technical problem of inaccurate task allocation in existing participant preference-based task allocation methods, which only consider a single dimension of user preferences without considering the user's subjective willingness to accept the task. The crowd-sensing task allocation method provided in this application fully utilizes the large amount of multimodal data generated by participants when completing crowd-sensing tasks. It models participants' preferences for tasks from three dimensions: content, time, and location. Then, it treats the task-participant matching degree prediction as a binary classification problem, using a Logit model to integrate preferences from different dimensions to predict the matching degree of each participant-task pair. Finally, the tasks to be assigned are distributed to the participants with the highest matching degrees. By accurately modeling participant preferences, the accuracy and efficiency of task allocation are improved.
[0005] To address the aforementioned technical problems, embodiments of this application provide a method for collective intelligent sensing task allocation based on the preferences of multiple participants, comprising the following steps: acquiring historical task data of several participants, wherein the historical task data includes text data, image data, time data, and location data for each task; constructing and acquiring a content preference model and the content preference of each participant based on the text data and image data, constructing and acquiring a time preference model and the time preference of each participant based on the time data, and constructing a location preference model based on the location data; receiving a new task, wherein the new task includes new text data, new image data, new time data, and new location data; inputting the new text data and new image data into the content preference model to obtain new content preferences, using the new content preferences and the content preferences of each participant to obtain the content matching degree of each participant, using the new time data and the time preferences of each participant to obtain the time matching degree of each participant, and inputting the new location data into the location preference model to obtain the location matching degree of each participant; weightedly fusing the content matching degree, time matching degree, and location matching degree to construct a task preference model for each participant; converting the task preference model of each participant into a matching probability between each participant and the new task; and allocating the new task based on the matching probability.
[0006] Embodiments of this application also provide a crowd-aware task allocation system based on the preferences of multiple participants, including: a data acquisition module, a single-dimensional preference acquisition module, a new task acceptance and processing module, a matching probability acquisition module, and a new task allocation module; the data acquisition module is used to acquire historical task data of several participants, wherein the historical task data includes text data, image data, time data, and location data of each task; the single-dimensional preference acquisition module is used to train and acquire a content preference model and the content preference of each participant based on the text data and image data, train and acquire a time preference model and the time preference of each participant based on the time data, and train and acquire a location preference model based on the location data; the new task acceptance and processing module is used to receive new tasks, wherein the new task... The system includes new text data, new image data, new time data, and new location data. The new text data and image data are input into a content preference model to obtain new content preferences. The content matching degree for each participant is obtained using the new content preferences and each participant's content preferences. The time matching degree for each participant is obtained using the new time data and each participant's time preferences. The location matching degree for each participant is obtained by inputting the new location data into a location preference model. The matching probability acquisition module is used to weightedly fuse the content matching degree, time matching degree, and location matching degree to construct a task preference model for each participant. The task preference model for each participant is then converted into a matching probability between each participant and the new task. The new task allocation module is used to allocate new tasks based on the matching probability.
[0007] Embodiments of this application also provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described crowd-aware task allocation method based on the preferences of multiple participants.
[0008] Embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described crowd-aware task allocation method based on the preferences of multiple participants.
[0009] The embodiments of this application provide a crowd-aware task allocation method based on the preferences of multiple participants. First, by guiding and fusing the features of complementary content such as images and text, the accuracy of deep feature extraction from users' historical task data is improved. Second, the embodiments of this application also integrate content information and spatiotemporal information from users' historical task data to characterize users' preferences for tasks, thereby improving the accuracy of task allocation. Finally, the embodiments of this application treat task allocation as a binary classification problem and use the Logit model to unify all factors into a single framework, calculating the matching probability between tasks and participants, thereby improving the efficiency of task allocation.
[0010] In some optional embodiments, the construction and acquisition of the content preference model and the content preferences of each participant based on text and image data includes:
[0011] A content preference model is trained and obtained based on text and image data. The content preferences of each participant are obtained by using the results of the content preference model based on all historical task inputs of each participant.
[0012] In some optional embodiments, the content preference model includes: a feature extraction module, a feature fusion module, and a multi-label classification module; the input of the content preference model is the text data and image data of the Nth task, and the output is the content preference of the Nth task.
[0013] In some optional embodiments, the feature fusion module includes:
[0014] Semantic alignment and cross-guidance are performed using text features and image features to generate guided image features and guided text features respectively. The guided image features and guided text features are then weighted and fused to construct the content features for each task.
[0015] In some optional embodiments, the content category of the Nth task is embedded as a tag into the multi-tag classification module.
[0016] In some alternative embodiments, the allocation of new tasks based on matching probabilities includes: assigning new tasks to a number of participants with the highest matching probabilities.
[0017] In some optional embodiments, after allocating new tasks based on matching probabilities, the method further includes: uploading the task allocation results to a platform, and the task publisher obtaining the task allocation results through the platform. Attached Figure Description
[0018] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.
[0019] Figure 1This is a flowchart of a crowd-sensing task allocation method based on the preferences of multiple participants provided in one embodiment of this application;
[0020] Figure 2 This is a framework diagram of a crowd-aware task allocation method based on the preferences of multiple participants provided in one embodiment of this application;
[0021] Figure 3 This is a framework diagram of the content preference extraction steps provided in one embodiment of this application;
[0022] Figure 4 This is a set of coordinates provided by one embodiment of this application when participants engage in a task;
[0023] Figure 5 This is a Gaussian mixture distribution diagram provided in one embodiment of this application;
[0024] Figure 6 This is another embodiment of the present application providing a schematic diagram of a crowd-aware task allocation system based on the preferences of multiple participants;
[0025] Figure 7 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the various embodiments of this application will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the various embodiments of this application to help readers better understand this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for the convenience of description and should not constitute any limitation on the specific implementation of this application. The various embodiments can be combined with and referenced by each other without contradiction.
[0027] One embodiment of this application relates to a crowd-sensing task allocation method based on the preferences of multiple participants, applied to an electronic device. The electronic device can be a terminal or a server. In this embodiment and the following embodiments, the electronic device is described using a server as an example. The implementation details of the crowd-sensing task allocation method based on the preferences of multiple participants in this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution.
[0028] The specific process of the crowd-sensing task allocation method based on the preferences of multiple participants in this embodiment can be described as follows: Figure 1 As shown, it includes:
[0029] Step 101: Obtain historical task data for several participants, including text data, image data, time data, and location data for each task.
[0030] It should be noted that the core focus of this application's embodiments is to assign tasks to the participants with the highest suitability. Therefore, the task allocation method based on multi-participant preferences proposed in this application first analyzes the historical task data generated when participants engage in tasks. This application analyzes participants' preferences for tasks from three dimensions: content, time, and location. By integrating preferences from different dimensions, the matching degree between different participants and tasks is predicted. Then, the task to be assigned is allocated to the participant with the highest matching degree, thus completing the crowd-aware perception task allocation based on multi-participant preferences.
[0031] Please see Figure 2 , Figure 2 A framework diagram of a task allocation method based on the preferences of multiple participants provided in an embodiment of this application is presented. The framework includes: preference extraction and task allocation, wherein preference extraction includes: participant content preference extraction, time preference extraction and location preference extraction.
[0032] In this embodiment, the task publisher releases the task content to the crowdsourcing sensing platform. The task content includes the task location, time, and details, and the task data is in text or image format. When the crowdsourcing sensing platform receives a new task, it analyzes the historical task data generated by several participants in the crowdsourcing sensing task, calculates the task preference of each participant, and then determines the matching degree between each participant and the new task based on the task preference.
[0033] In this embodiment, T is denoted as {t1, t2, ..., t}. n} represents the set of tasks in the platform, where t i Let T represent the i-th task. u This represents the historical task data of the u-th participant. The historical task data includes text data, image data, location data, and time data.
[0034] Step 102: Train and obtain the content preference model and the content preference of each participant based on text data and image data; train and obtain the time preference model and the time preference of each participant based on time data; train and obtain the location preference model based on location data.
[0035] This embodiment is recorded as TP. u LP u CP u Let represent the time preference, location preference, and content preference of the u-th participant for the task.
[0036] This embodiment first constructs and obtains a content preference model and the content preferences of each participant based on text data and image data.
[0037] Please see Figure 3 , Figure 3 The overall framework diagram of content preference extraction in this embodiment, i.e., the overall architecture diagram of the content preference model, is presented. This framework mainly includes three parts: a feature extraction module, a feature fusion module, and a multi-label classification module. The framework execution flow is as follows: First, features of text and images are extracted separately. Then, the text features and image features are fused using a cross-attention method. Next, a Transformer decoder is used to perform multi-label classification to obtain the category labels of the data. Finally, the category labels of all historical task data of the participant are merged to obtain the participant's preference for task content. The specific steps are as follows:
[0038] First, the text and image data from the historical task data are preprocessed to obtain text and image features. The specific preprocessing steps are as follows: The pre-trained BERT model is used to extract text features, resulting in text features FT = BERT(t). Then, based on the Transformer model, a CNN model is integrated, and the outputs of the two models are concatenated, giving the model both local and global feature extraction capabilities, thus obtaining image features FI = Transformer(t) + CNN(t).
[0039] This completes the preprocessing of text and image data from each participant's historical task data, extracting text and image features.
[0040] It should be noted that in collective intelligence tasks, the data of various modalities are often very concise, and the data of any modality is equally important to the task. Therefore, the embodiments of this application consider using text features and image features for semantic alignment and cross-guidance to form text-guided attention features and image-guided attention features respectively, so as to achieve complementarity and enhancement of the two modalities.
[0041] This application uses a feature fusion method based on cross-attention mechanism to achieve complementarity and enhancement of two modalities. Specifically, when text guides image features, text is used to enhance features in the image features that the text also focuses on. When image guides text features, image is used to enhance features in the text features that the image also focuses on.
[0042] The specific steps for fusing image features and text features based on the cross-attention mechanism in this embodiment are as follows:
[0043] Let the text features be FT = {α1, α2, ..., α} n The image features are FI = {η1, η2, ..., η}.n In text-guided image feature generation, the text is treated as a query (Q), and the image features are treated as keys (K) and values (V), i.e. The guided image features are:
[0044]
[0045] Where d is the dimension of the key vector.
[0046] When using image-guided text features, the image features are treated as a query (Q), and the text features are treated as keys (K) and values (V), i.e. The text features after guidance are:
[0047]
[0048] Where d is the dimension of the key vector.
[0049] The guided image features FI′ and the guided text features FT′ are fused to obtain the content feature F. The fusion calculation formula is as follows:
[0050] F=λFI′+(1-λ)FT′
[0051] Where λ represents the weight of the image features. It should be noted that in this embodiment, λ is set to 0.39, but implementers can adjust it according to the specific implementation scenario.
[0052] At this point, feature fusion is performed on image features and text features to obtain the content features of each task in each participant's historical task data.
[0053] It should be noted that after the above processing, the content features of each task in each participant's historical task data are obtained. Then, a multi-label classification process is performed. In this embodiment, a Transformer-based decoder is used to implement multi-label classification. Compared to traditional Transformer decoders that require pre-setting labels for neural network training, this embodiment converts all category labels in the data into label embeddings, which are then used as queries for the decoder to check the existence of each label and execute a multi-head attention mechanism to strengthen category-related features.
[0054] The specific steps for implementing multi-label classification using a Transformer decoder in this embodiment are as follows:
[0055] This embodiment takes the Nth task in the historical task data of the Hth participant as an example. The category of all content in the Nth task is used as the label. The label embedding Q passes through a self-attention layer to learn the correlation between labels. Then, it enters a multi-head cross-attention layer with the feature vector extracted by the feature extraction layer, which is the content feature, to strengthen the category-related features. Finally, it enters a feedforward neural network layer to perform linear processing on the features. The input of each Transformer decoder layer i comes from the (i-1)th decoder layer. The outputs of each layer in the decoder are as follows:
[0056]
[0057] O i =FFN(O′ i )
[0058] in:
[0059] MultiHead(Q,K,V)=Concat(h1,h2,…,h n W p
[0060]
[0061] FFN(x) = max(0, xW1+b1)W2+b2
[0062] W represents different transformation matrices.
[0063] To reduce the impact of label imbalance on the results, this embodiment uses focus loss as the loss function, which allows the model to pay more attention to samples that are difficult to classify and improves the ability to identify minority categories.
[0064]
[0065] The final output of the model is: O j = [o1, o2, ..., o h ], where h is the number of tag categories, o i Let be the predicted probability of the i-th class among the features.
[0066] This completes the construction of the content preference model.
[0067] The above processing was performed on the text and image data of each task in all of the participant's historical task data, and the output results were combined to obtain the participant's content preference (CP). u .
[0068] CP u =Concat(O j )
[0069] It should be noted that this embodiment obtains all outputs O from all historical task data of the participants. j Summing the results and dividing by the total number of categories yields the participant's content preferences.
[0070] At this point, the content preferences of each participant were obtained.
[0071] Furthermore, this embodiment constructs and obtains a time preference model and the time preferences of each participant based on time data.
[0072] Using a 30-dimensional vector D g This represents the task execution time within a month. If the task execution time falls within a specific time period, the corresponding position in the vector is set to 1, and the rest are set to 0. In this embodiment, a month is defined as 30 days. Therefore, a time preference model is constructed, with the execution time of a task as the input and the time preference for that task as the output. The time preference TP of the u-th participant for the task is used as the starting point. u For example, the specific calculation formula for the time preference model is as follows:
[0073]
[0074] Among them, t i Let T represent the i-th task. u This represents the historical task data of the u-th participant.
[0075] Thus, we have obtained the time preference model and the time preferences of each participant.
[0076] Furthermore, this embodiment constructs a location preference model based on location data.
[0077] This embodiment uses a Gaussian mixture distribution-based method to model the location data and obtain the location preferences of each participant. The input to the location preference model is the task's location data, and the output is the participant's location preference for the task. It should be noted that the location data is the set of coordinates of the participant when participating in the task. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 The set of coordinates of several participants when they participate in a task is presented. Let P be the set of coordinates of the u-th participant when they choose a task. u for:
[0078]
[0079] in, Represents the i-th task t i The coordinates of T u This represents the historical task data of the u-th participant.
[0080] The Gaussian mixture distribution of the historical task locations for the u-th participant is:
[0081]
[0082] in:
[0083]
[0084] In the formula, D represents the dimension of the historical task data of the u-th participant. It should be noted that this formula is a Gaussian mixture distribution formula, and the meanings of the letters in this formula are consistent with those in the Gaussian mixture distribution formula.
[0085] Furthermore, participants' location preferences for tasks (LP) u = p(x).
[0086] Please see Figure 5 , Figure 5 The Gaussian mixture distribution obtained in this embodiment is shown. It should be noted that when a new task arises, the degree of location matching between the task and the participants can be calculated based on the Gaussian mixture distribution. As can be seen from the figure, the darker the color, the more the task's scope matches the participants' activity patterns.
[0087] This completes the construction of the location preference model.
[0088] Step 103: Receive the new task and obtain the content matching degree, time matching degree, and location matching degree based on the new task, content preference model, time preference model, and location preference model.
[0089] It should be noted that after obtaining each participant's preferences for the task across three dimensions, when a new task appears on the platform, the matching degree between each participant and the task across each dimension can be calculated.
[0090] The system receives new tasks from the platform, including new text data, new image data, new time data, and new location data. The new text data and new image data are input into a content preference model to obtain new content preferences. The new content preferences and each participant's content preferences are used to obtain the content matching degree for each participant. The new time data and each participant's time preferences are used to obtain the time matching degree for each participant. The new location data is input into a location preference model to obtain the location matching degree for each participant.
[0091] Let the new task be A. Then the content matching degree of the u-th participant with the new task A is: The time matching score and location matching score are respectively: Where cos(*) denotes the cosine similarity, O A To obtain the output of the content preference model after inputting the new task A, D A For the time data of the new task A, i.e., the execution time, l AThe location data, i.e., coordinates, for the new task.
[0092] At this point, we have obtained the content matching degree, time matching degree, and location matching degree for each participant.
[0093] Step 104: Weighted fusion of content matching degree, time matching degree and location matching degree to construct a task preference model for each participant; transform the task preference model of each participant into the matching probability between each participant and the new task.
[0094] This embodiment takes the u-th participant as an example, and the specific formula for constructing the preference model of the u-th participant is as follows:
[0095]
[0096] Where W = (w1, w2, w3) represents the weight of different features, and b represents the bias. This indicates the degree to which the u-th participant matches the content of the new task A. This indicates the degree of time compatibility between the u-th participant and the new task A. This indicates the degree to which the u-th participant matches the location of the new task A.
[0097] Since the outcome of a participant's acceptance or rejection of a task is binary, this paper transforms the prediction of the participant-task match into a binary classification problem. This paper uses a logistic function to convert the preference model into a probability function, that is, to transform the preference model into a matching probability, as follows:
[0098]
[0099] Let the cost function be:
[0100]
[0101] At this point, the matching probability of each participant and the new task is obtained.
[0102] Step 105: Assign new tasks based on matching probabilities.
[0103] After the platform receives a new task, it calculates the matching probability between each participant and the task, and assigns the task to the participants with the highest matching probability according to the task requirements.
[0104] After receiving the assigned task, participants can accept or reject the task according to their own wishes. If they accept the task, they should complete the task as required and upload the results to the platform.
[0105] The task publisher obtains the task allocation results through the platform.
[0106] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.
[0107] Another embodiment of this application relates to a crowd-sensing task allocation system based on the preferences of multiple participants. The implementation details of this embodiment's crowd-sensing task allocation system are described below. The following details are provided for ease of understanding and are not essential for implementing this solution. A schematic diagram of this embodiment's crowd-sensing task allocation system based on the preferences of multiple participants can be seen as follows: Figure 6 As shown, it includes:
[0108] Data acquisition module 201, single-dimensional preference acquisition module 202, new task acceptance and processing module 203, matching probability acquisition module 204, new task allocation module 205;
[0109] The data acquisition module 201 is used to acquire historical task data of several participants, wherein the historical task data includes text data, image data, time data and location data of each task;
[0110] The single-dimensional preference acquisition module 202 is used to train and acquire a content preference model and the content preference of each participant based on text data and image data, train and acquire a time preference model and the time preference of each participant based on time data, and train and acquire a location preference model based on location data.
[0111] The new task receiving and processing module 203 is used to receive new tasks and obtain the content matching degree, time matching degree and location matching degree based on the new task, content preference model, time preference model and location preference model;
[0112] The matching probability acquisition module 204 is used to construct a task preference model for each participant by weighted fusion of content matching degree, time matching degree and location matching degree; and to convert the task preference model of each participant into the matching probability between each participant and the new task.
[0113] The new task allocation module 205 is used to allocate new tasks based on matching probability.
[0114] It is not difficult to see that this embodiment is a system embodiment corresponding to the above method embodiments, and this embodiment can be implemented in conjunction with the above method embodiments. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.
[0115] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent in this embodiment.
[0116] Another embodiment of this application relates to an electronic device, such as... Figure 7 As shown, it includes: at least one processor 301; and a memory 302 communicatively connected to the at least one processor 301; wherein the memory 302 stores instructions executable by the at least one processor 301, the instructions being executed by the at least one processor 301 to enable the at least one processor 301 to execute the crowd intelligence perception task allocation method based on the preferences of multiple participants in the above embodiments.
[0117] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.
[0118] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.
[0119] Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.
[0120] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0121] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.
Claims
1. A method for allocating crowd-sensing tasks based on the preferences of multiple participants, characterized in that, include: Acquire historical task data from several participants, including text data, image data, time data, and location data; A content preference model and each participant's content preferences are constructed and obtained based on text and image data; a time preference model and each participant's time preferences are constructed and obtained based on time data; and a location preference model is constructed based on location data. Receive new tasks, which include new text data, new image data, new time data, and new location data; New text data and new image data are input into the content preference model to obtain new content preferences. The content matching degree of each participant is obtained by using the new content preferences and the content preferences of each participant. The time matching degree of each participant is obtained by using the new time data and the time preferences of each participant. The location matching degree of each participant is obtained by inputting the new location data into the location preference model. A task preference model for each participant is constructed by weighting and fusing the content matching degree, time matching degree, and location matching degree. Transform each participant's task preference model into the matching probability between each participant and the new task; New tasks are assigned based on matching probability; The process of constructing and obtaining a content preference model and the content preferences of each participant based on text and image data includes: A content preference model is constructed based on text and image data. The content preference of each participant is obtained by using the results of the content preference model with all of their historical task inputs. The content preference model includes: Feature extraction module, feature fusion module, and multi-label classification module; The input to the content preference model is the first... The text and image data for the task are output as the first task. Content preferences for each task; The feature fusion module includes: Semantic alignment and cross-guidance are performed using text features and image features to form guided image features and guided text features respectively. The guided image features and guided text features are then weighted and fused to construct the content features for each task. The first The content category of each task is embedded as a tag in the multi-tag classification module.
2. The crowd-based intelligent perception task allocation method based on the preferences of multiple participants as described in claim 1, characterized in that, The allocation of new tasks based on matching probability includes: assigning new tasks to several participants with the highest matching probability.
3. The crowd-based intelligent perception task allocation method based on the preferences of multiple participants according to claim 1, characterized in that, After allocating new tasks based on matching probabilities, the method further includes: uploading the task allocation results to the platform, and the task publisher obtaining the task allocation results through the platform.
4. A crowd-based intelligent sensing task allocation system based on the preferences of multiple participants, characterized in that: The system includes a data acquisition module, a single-dimensional preference acquisition module, a new task acceptance and processing module, a matching probability acquisition module, and a new task allocation module. The data acquisition module is used to acquire historical task data of several participants, wherein the historical task data includes text data, image data, time data and location data of each task; The single-dimensional preference acquisition module is used to train and acquire a content preference model and the content preference of each participant based on text data and image data, to train and acquire a time preference model and the time preference of each participant based on time data, and to train and acquire a location preference model based on location data. The new task receiving and processing module is used to receive new tasks, which include new text data, new image data, new time data, and new location data. The new text data and new image data are input into the content preference model to obtain new content preferences. The content matching degree of each participant is obtained by using the new content preferences and the content preferences of each participant. The time matching degree of each participant is obtained by using the new time data and the time preferences of each participant. The location data is input into the location preference model to obtain the location matching degree of each participant. The matching probability acquisition module is used to weight and fuse the content matching degree, time matching degree, and location matching degree to construct a task preference model for each participant; and to convert the task preference model of each participant into the matching probability between each participant and the new task. The new task allocation module is used to allocate new tasks based on matching probability; The process of constructing and obtaining a content preference model and the content preferences of each participant based on text and image data includes: A content preference model is constructed based on text and image data. The content preference of each participant is obtained by using the results of the content preference model with all of their historical task inputs. The content preference model includes: Feature extraction module, feature fusion module, and multi-label classification module; The input to the content preference model is the first... The text and image data for the task are output as the first task. Content preferences for each task; The feature fusion module includes: Semantic alignment and cross-guidance are performed using text features and image features to form guided image features and guided text features respectively. The guided image features and guided text features are then weighted and fused to construct the content features for each task. The first The content category of each task is embedded as a tag in the multi-tag classification module.
5. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the crowd-aware task allocation method based on the preferences of multiple participants as described in any one of claims 1 to 3.
6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the crowd-aware task allocation method based on the preferences of multiple participants as described in any one of claims 1 to 3.