Machine Learning-Based Personalized Recommendation Method and System for Crowdsourcing Tasks

Abstract

The present invention provides a method and system for personalized recommendation of crowdsourcing tasks based on machine learning, which relates to the technical field of task scheduling. According to the existing literature on crowdsourcing competitions and the available information on crowdsourcing platforms, the present invention constructs a complete and detailed worker feature recognition system based on the Fogg behavior model that integrates motivation and ability theory. At the same time, by crawling the historical data of workers and the task data for a period of time, the present worker-task interaction data is constructed to form a worker dataset covering the past and present information of workers, so as to better measure worker features. It helps the platform quickly identify workers with high value, generates a recommendation list based on the selection probability, and finally realizes the personalized recommendation of crowdsourcing tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of task scheduling, and particularly to a personalized recommendation method and system for crowdsourcing tasks based on machine learning. Background Art

[0002] Crowdsourcing refers to the practice of a company or organization outsourcing work tasks previously performed by employees to non-specific volunteer members of the public in a free and voluntary manner. "Crowdsourcing is socialized production", where users jointly create value, which is a typical application scenario of collective wisdom collaboration, and can greatly improve the efficiency of task completion and give full play to the collective role in a group society.

[0003] In traditional task recommendation methods, the method for recommending to users is as follows: First, based on task data and worker data on a crowdsourcing platform, a feature vector table of worker capabilities is obtained and output. Then, a task recommendation list is generated based on the task bidding records of workers with similar capabilities and the bidding records of target workers for recommendation.

[0004] However, the existing recommendation algorithms have the following defects:

[0005] 1. The measurement criteria for worker characteristics are relatively rough, often identifying from a single dimension such as motivation or ability, or failing to completely identify the determinants or relevant predictors of characteristics based on relevant theories;

[0006] 2. Traditional recommendations often only rely on historical data of tasks and workers, while the interaction characteristics between workers and tasks change over time, resulting in limited acquisition of worker characteristic data and unable to effectively measure worker characteristics;

[0007] 3. Currently, the recommendation succession is from feature modeling of a single attribute, to joint interaction between users and tasks, and then to obtaining implicit feedback information in features. Most still remain in the modeling of single attributes and joint interactions, and a small number have begun to focus on implicit feedback information. Summary of the Invention

[0008] (1) Technical Problems to be Solved

[0009] In view of the deficiencies of the prior art, the present invention provides a personalized recommendation method and system for crowdsourcing tasks based on machine learning, which solves the problem of poor accuracy in identifying worker characteristics in current methods.

[0010] (2) Technical Solutions

[0011] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0012] In the first aspect, a personalized recommendation method for crowdsourcing tasks based on machine learning is provided. The method includes:

[0013] Obtain the historical feature data of workers and the task feature data; and obtain the worker-task interaction feature data based on the task feature data;

[0014] Train a worker classification model considering worker motivation and worker ability based on the preprocessed historical feature data of workers and the worker-task interaction feature data.

[0015] Based on the trained worker classification model, select workers with high motivation and high ability from the worker feature dataset to be classified, and recommend tasks to them.

[0016] Furthermore, the obtaining of the historical feature data of workers and the task feature data; and the obtaining of the worker-task interaction feature data based on the task feature data includes:

[0017] S101. Crawl all the historical feature data of workers and the task feature data within a period of time from the crowdsourcing platform;

[0018] S102. Preprocess the task feature data and encode the task feature data to obtain a task feature dataset;

[0019] S103. Divide the task feature dataset into a training set and a test set in a 1:1 ratio according to the time sequence, and obtain the worker-task interaction feature data based on the training set;

[0020] S104. Preprocess the historical feature data of workers and encode it to obtain a historical feature dataset of workers;

[0021] S105. Preprocess the worker-task interaction feature data and encode it.

[0022] Furthermore, the preprocessing of the task feature data includes:

[0023] Screen out the task feature data with the number of task bidders greater than or equal to M1;

[0024] Eliminate tasks with uncompleted recruitment or non-competition tasks;

[0025] Eliminate tasks with missing key features such as the elected reward of the task, the task release time, and the task deadline;

[0026] Deduplicate the duplicate task feature data in the list;

[0027] Eliminate outliers in the task list;

[0028] The preprocessing of the historical feature data of workers includes:

[0029] Eliminate workers with the number of past elected suggestions less than or equal to M2;

[0030] Eliminate workers whose past recommended task numbers are less than or equal to M3;

[0031] The preprocessing of the worker-task interaction feature data includes:

[0032] Eliminate workers whose number of participated tasks is less than M4;

[0033] Eliminate workers whose number of won tasks is less than M5;

[0034] Complete the missing values.

[0035] Furthermore, training a worker classification model considering worker motivation and worker ability based on the preprocessed worker historical feature data and worker-task interaction feature data includes:

[0036] S201. Construct the ability index and motivation index of workers;

[0037] S202. Determine the weights of the variables of the ability index of workers and the weights of the variables of the motivation index, and obtain the motivation value and ability value of each worker;

[0038] S203. Obtain a threshold line model for use as a criterion for dividing high-motivation and high-ability workers.

[0039] Furthermore, the ability index of the workers includes: inherent ability, professional ability, general experience, professional experience, diversity, complexity;

[0040] The motivation index of the workers includes: pleasure and fun, work autonomy, task complexity, self-marketing / sense of belonging;

[0041] And the variables of the inherent ability include: platform rating, platform certification level, past number of won bids;

[0042] The variables of the professional ability include: number of certification categories, number of certification sub-categories, number of certification skills, number of skills over ten years;

[0043] The variables of the general experience include: past number of task recommendations, recent number of participated tasks, recent number of won tasks, hourly rate;

[0044] The variables of the professional experience include: number of certified industries, number of tasks included in recent participated tasks, number of tasks included in recent won tasks;

[0045] The variables of the diversity include: number of recent participated different task categories, number of recent participated different task sub-categories, number of recent won different task categories, number of recent won different task sub-categories;

[0046] The variables of the complexity include: the average number of bidders participating in tasks recently, the average number of suggestions for tasks recently, the average number of collectors for tasks recently, the average number of views for tasks recently;

[0047] The variables of the enjoyment and fun include: the average task reward, the task approval rate;

[0048] The variables of the work autonomy include: the average recruitment time, the task completion rate;

[0049] The variables of the task complexity include: the ratio of the average number of collectors to the number of bidders for tasks participated, the ratio of the average number of collectors to the number of bidders for winning tasks, the ratio of the average number of viewers to the number of collectors for tasks participated, the ratio of the average number of viewers to the number of collectors for winning tasks;

[0050] The variables of the self - marketing / sense of belonging include: the favorable comment rate, the number of comments.

[0051] Furthermore, determining the weights of the variables of the ability index and the weights of the variables of the motivation index for workers to obtain the motivation value and ability value of each worker includes:

[0052] Determining the weights of each variable based on the CRITIC objective weighting method;

[0053] And the calculation formulas for the motivation value and ability value of workers are as follows:

[0054]

[0055]

[0056] a i represents the magnitude of the ability value of the i - th worker;

[0057] m i represents the magnitude of the motivation value of the i - th worker;

[0058] represents the magnitude of the objective weight of the j - th ability variable;

[0059] represents the magnitude of the objective weight of the k - th motivation variable;

[0060] x ij represents the value of the j - th ability variable of the i - th worker;

[0061] y ik represents the value of the k - th motivation variable of the i - th worker.

[0062] Furthermore, the threshold line model includes:

[0063] Motivition = mi -M f +β

[0064] Ability = a i -A f +α

[0065] Wherein,

[0066] M f is the Kolmogorov average value of the worker motivation;

[0067] A f is the Kolmogorov average value of the worker ability;

[0068] α represents the parameter adjusted on the ability axis;

[0069] β represents the parameter adjusted on the motivation axis;

[0070] And by constructing a loss function lost, using five-fold cross-validation to train α and β to make lost converge, and taking the average value of α and β as the final result.

[0071] Furthermore, based on the trained worker classification model, screening out workers with high motivation and high ability from the worker feature dataset to be classified includes: when the motivation value and ability value of the worker to be classified are both greater than the threshold line, it is identified as a worker with high motivation and high ability.

[0072] In a second aspect, a crowdsourcing task personalized recommendation system based on machine learning is provided. The system includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned method are implemented.

[0073] (III) Beneficial effects

[0074] The present invention provides a crowdsourcing task personalized recommendation method and system based on machine learning. Compared with the prior art, it has the following beneficial effects:

[0075] Based on the existing literature of crowdsourcing competitions and the available information on crowdsourcing platforms, the present invention constructs a complete and refined worker feature recognition system based on the Fogg behavior model that integrates motivation and ability theories. At the same time, by crawling the historical data of workers and the task data for a period of time, the current worker-task interaction data is constructed to form a worker dataset covering the past and present information of workers, so as to better measure worker features. It helps the platform quickly identify workers with high value, generates a recommendation list based on the selection probability, and finally realizes the personalized recommendation of crowdsourcing tasks. Description of the drawings

[0076] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0077] Figure 1 It is a flowchart of an embodiment of the present invention. Detailed implementation manners

[0078] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0079] The embodiments of the present application provide a personalized recommendation method and system for crowdsourcing tasks based on machine learning, which solves the problem of poor accuracy in identifying worker characteristics in current methods.

[0080] To better understand the above technical solutions, the following will describe the above technical solutions in detail in conjunction with the accompanying drawings of the specification and specific implementation manners.

[0081] Embodiment 1:

[0082] As Figure 1 shown, the present invention provides a personalized recommendation method for crowdsourcing tasks based on machine learning, and the method includes:

[0083] S1. Obtain the historical feature data of workers and the task feature data; and obtain the worker-task interaction feature data based on the task feature data;

[0084] S2. Train a worker classification model considering worker motivation and worker ability based on the preprocessed historical feature data of workers and the worker-task interaction feature data.

[0085] S3. Based on the trained worker classification model, screen out workers with high motivation and high ability from the dataset of worker feature data to be classified, and recommend tasks to them.

[0086] The beneficial effects of this embodiment are:

[0087] Based on the existing literature on crowdsourcing competitions and the available information on crowdsourcing platforms, this invention constructs a complete and detailed worker feature recognition system based on the Fogg Behavior Model that integrates motivation and ability theory. At the same time, by scraping the historical data of workers and the task data for a period of time, the current worker-task interaction data is constructed to form a worker dataset covering the past and present information of workers, so as to better measure worker features. It helps the platform quickly identify high-value workers, generate a recommendation list based on the selection probability, and finally achieve personalized recommendations for crowdsourcing tasks.

[0088] The implementation process of the embodiments of this invention will be described in detail below:

[0089] S1. Obtain the historical feature data of workers and the task feature data; and obtain the worker-task interaction feature data based on the task feature data.

[0090] Specifically, it includes the following steps:

[0091] S101. Scrape all the historical feature data of workers and the task feature data for a period of time from the crowdsourcing platform.

[0092] In specific implementation, a lightweight crawler framework scrapy based on Python can be used. Through XPath and CSS expressions for web page data extraction and parsing, the Redis database is used as a distributed shared crawler queue, and the MongoDB database is used as a data storage repository. The Selenium automation testing tool is integrated, and middleware such as random User-Agent, Alibaba Cloud proxy IP, and self-built proxy IP pool are used, and it is deployed to a cloud server to scrape the historical feature data of all workers in a certain category and achieve large-scale real-time incremental scraping of the historical feature data of workers and the task feature data on the crowdsourcing platform.

[0093] And as shown in Table 1, the historical feature data of workers includes: worker ID, platform authentication level, platform score, worker positive review rate, task completion rate, task approval rate, hourly work rate, authentication category, authentication industry, authentication skills, skill duration, total number of past suggestions (total number of task suggestion solutions submitted in the past), total number of past elected suggestions (number of task solution rankings first).

[0094] The task feature data includes: task name, task recruitment status, task type, task category, task sub-category, task industry, task release time, task deadline, task winning reward (monetary reward obtained for ranking first in the task solution), task participation reward (varies by task. For some tasks to attract more workers to bid, the task reward is divided into winning reward and participation reward. In addition to giving most of the bonus to the first place, a small part of the bonus is also provided to the top few except the first place), task suggestion count (how many task suggestion solutions are submitted by workers for this task), number of task bidders, number of task winners, number of task favorites, and number of task views. In the present invention, winning the bid means being elected or a parameter.

[0095] Table 1

[0096]

[0097]

[0098] S102. After obtaining the crawled data, first preprocess the task feature data and encode the task feature data to obtain a task feature data set.

[0099] Specifically, the preprocessing mainly includes data screening and data cleaning in sequence.

[0100] Among them, the data screening includes:

[0101] 1) Screen out task feature data with the number of task bidders greater than or equal to M1.

[0102] 2) Exclude tasks with recruitment not ended or non-competition tasks. (This embodiment is only for competition tasks).

[0103] Project tasks: The task bonus is an interval, the submitted suggestion is a proposal, and a contract needs to be signed for subsequent completion.

[0104] Competition tasks: The task bonus is a fixed value, and the bonus can be obtained if the submitted suggestion wins the bid.

[0105] Task tasks: The bonus can be obtained by completing the task (such as filling out a questionnaire).

[0106] 3) Exclude tasks with missing key features such as task winning reward, task release time, and task deadline.

[0107] The data cleaning includes:

[0108] 1) Remove duplicates from the repeated task feature data in the list;

[0109] 2) Exclude outliers from the task list.

[0110] M1 is a preset threshold value, which can be set according to actual needs.

[0111] For data encoding, in specific implementation, a Python running environment can be constructed first, and corresponding encoding methods can be adopted for each preprocessed data.

[0112] For the classified and graded data, label encoding is adopted so that each type can be described by a number;

[0113] For the data with specific numerical values, direct encoding is performed;

[0114] It is also necessary to perform normalization processing on it, because the types and dimensions of the sample feature data are different, and the absolute values vary greatly, which will cause some features with smaller value ranges to be ignored. At the same time, normalizing the data can improve the convergence speed and accuracy of the model.

[0115] So far, the task feature data set can be obtained.

[0116] S103. Divide the task feature data set into a training set and a test set in a 1:1 ratio according to the time sequence, and use the task training set to obtain worker-task interaction feature data through the pandas and numpy packages of Python.

[0117] As shown in Table 2, the worker-task interaction feature data includes: the name of the tasks participated by each worker, the number of tasks participated, the category of tasks participated, the number of task categories participated, the sub-category of tasks participated, the number of sub-categories of tasks participated, the industry of tasks participated, the number of task industries participated, the name of the winning tasks, the number of winning tasks, the category of winning tasks, the sub-category of winning tasks, the number of sub-categories of winning tasks, the industry of winning tasks, the number of task industries of winning tasks, the average reward for participating tasks, the average working time for participating tasks, the average reward for winning tasks, and the average working time for winning tasks.

[0118] Table 2

[0119]

[0120]

[0121] S104. Preprocess the crawled historical feature data of workers and perform encoding to obtain the historical feature data set of workers.

[0122] In specific implementation, the preprocessing mainly includes data screening and data cleaning in sequence.

[0123] Among them, the data screening includes:

[0124] Eliminate the following historical feature data of workers:

[0125] 1) Workers whose past elected proposal count is less than or equal to M2.

[0126] 2) Workers whose past proposal count is less than or equal to M3.

[0127] For the encoding of workers' historical feature data, in specific implementation, a Python running environment can be constructed first, and corresponding encoding methods can be adopted for each preprocessed data.

[0128] For the classified and graded data, label encoding is adopted so that each type can be described by a number;

[0129] For the data with specific numerical values, direct encoding is performed;

[0130] It is also necessary to perform normalization processing on it, because the types and dimensions of the sample feature data are different, and the absolute values vary greatly, which will cause some features with a smaller value range to be ignored. At the same time, normalizing the data can improve the convergence speed and accuracy of the model.

[0131] So far, the workers' historical feature dataset can be obtained. For the division of the workers' historical feature dataset, the following method can be adopted:

[0132] The workers' historical feature dataset is divided into a training set and a test set according to 8:2, and then the k-fold cross-validation is used to divide the workers' training set into a training set and a validation set.

[0133] S105. Preprocess the worker-task interaction feature data and perform encoding.

[0134] In specific implementation, the preprocessing mainly includes data screening and data cleaning in sequence.

[0135] Among them, the data screening includes:

[0136] Eliminate the following worker-task interaction feature data:

[0137] 1) Workers with less than M4 tasks participated;

[0138] 2) Workers with less than M5 winning tasks;

[0139] M2 to M5 are preset thresholds and can be set according to actual needs.

[0140] The data cleaning includes:

[0141] Complete the missing values. In this embodiment, there is no limitation on the method for completing the missing values. For example, based on the worker-task interaction feature data, the similarity between workers can be calculated by the Jaccard similarity method, that is, calculating the proportion of the number of intersection elements of any two workers A and worker B in the union of A and B, which is the Jaccard similarity between worker A and worker B. Calculate the Jaccard similarity between worker A and each worker, and the value obtained by weighted averaging is the supplementary missing value.

[0142] For the encoding operation, the same encoding method for the worker historical feature data and the task feature data can be adopted.

[0143] S2. Train a worker classification model considering worker motivation and worker ability based on the preprocessed worker historical feature data and worker-task interaction feature data.

[0144] Specifically, it includes the following steps:

[0145] S201. Construct the ability index and motivation index of workers.

[0146] For worker ability, the ability index of workers is extracted from the six indexes of inherent ability, professional ability, general experience, professional experience, diversity, and complexity for worker characteristics and worker-task interaction characteristics.

[0147] Among them, the inherent ability is determined according to variables x1 to x3;

[0148] The professional ability is determined according to variables x4 to x7;

[0149] The general experience is determined according to variables x8 to x 11 Determine;

[0150] The professional experience is determined according to variables x 12 ~x 14 Determine;

[0151] The diversity is determined according to variables x 15 ~x 18 Determine;

[0152] The complexity is determined according to variables x 19 ~x 22 Determine;

[0153] For worker motivation, the motivation index of workers is extracted from the four indexes of pleasure and fun, work autonomy, task complexity, and self-marketing / sense of belonging.

[0154] Among them, the pleasure and fun are determined according to variables y1 to y2;

[0155] The work autonomy is determined according to variables y3 to y4;

[0156] The task complexity is determined according to variables x5 to x8;

[0157] Self - marketing / sense of belonging is determined according to variables x9 to x 10 Determined.

[0158] The predictive factors of workers' motivation and ability and the related variables are shown in Table 3:

[0159] Table 3

[0160]

[0161]

[0162] S202. Determine the weights of the variables of the ability index and the weights of the variables of the motivation index of the workers, and obtain the motivation value and ability value of each worker.

[0163] In specific implementation, the CRITIC objective weighting method can be used to comprehensively measure the objective weights of the variables.

[0164] CRITIC objective weighting method: Based on the comparison intensity of the evaluation variables and the conflict between variables to comprehensively measure the objective weights of the variables. While considering the magnitude of the variability of the variables, the correlation between variables is also taken into account. It is not that the larger the number, the more important it is. It fully utilizes the objective attributes of the data itself for scientific evaluation.

[0165] Among them, the comparison intensity refers to the size of the value gap between the evaluation schemes of the same variable, which is represented by the standard deviation. The larger the standard deviation, the higher the weight. And in this embodiment:

[0166] x ij represents the value of the j - th ability variable of the i - th worker;

[0167] represents the mean value of the j - th ability variable;

[0168] represents the standard deviation of the j - th ability variable;

[0169] y ik represents the value of the k - th motivation variable of the i - th worker;

[0170] represents the mean value of the k - th motivation variable;

[0171] represents the standard deviation of the k - th motivation variable.

[0172] And the conflict between variables means that if there is a strong positive correlation between two variables, it indicates that their conflict is smaller and the weight will be lower, which is represented by the correlation coefficient. And in this implementation:

[0173] r ij The correlation coefficient between the i-th variable and the j-th variable representing ability, and its conflictiveness is expressed as (1 - r ij );

[0174] Represents the sum of the conflictiveness between the j-th variable of ability and other variables;

[0175] Represents the sum of the conflictiveness between the k-th variable of motivation and other variables.

[0176] And the calculation formulas for the above parameters are as follows:

[0177]

[0178]

[0179]

[0180]

[0181]

[0182]

[0183] And the amount of information contained in a variable is represented by the product of the comparison intensity and the conflictiveness of the variable:

[0184]

[0185]

[0186] Therefore, the calculation formula for the objective weight of a variable is as follows:

[0187]

[0188]

[0189] The calculation formulas for the motivation value and ability value of a worker are as follows:

[0190]

[0191]

[0192] a i Represents the actual ability value of the i-th worker;

[0193] m i Represents the actual motivation value of the i-th worker;

[0194] p represents the number of ability variables;

[0195] q represents the number of motivation variables;

[0196] represents the amount of information contained in the j-th ability variable;

[0197] represents the objective weight of the j-th ability variable;

[0198] represents the amount of information contained in the k-th motivation variable;

[0199] represents the objective weight of the k-th motivation variable.

[0200] S203. Construct a threshold line model to be used as a criterion for dividing workers with high motivation and high ability.

[0201] The threshold line consists of a theoretical limit, which indicates that for values above this limit, it shows sufficient high motivation and ability, and for the behavior change of this worker, only a signal type trigger is required, and the correct positioning of this limit is intrinsically related to the worker's dataset.

[0202] Based on the research conceptually described by the Fogg Behavior Model, define its fixed threshold line mathematical model as:

[0203]

[0204] The central tendency of the dataset relocates the threshold line based on the Kolmogorov mean value, and at the same time adopts the threshold line displacement standard, so as to better implement the trigger and identify workers with high motivation and high ability.

[0205] Among them

[0206] M f is the Kolmogorov mean value of the worker's motivation,

[0207] A f is the Kolmogorov mean value of the worker's ability,

[0208] The specific calculation formula is as follows:

[0209]

[0210]

[0211] Considering the equation in the application domain, α and β are added to motivation and ability respectively to adjust the displacement of the threshold line. By constructing a loss function lost, α and β are trained using five-fold cross-validation to make lost converge, and the average values of α and β are taken as the training results. The specific calculation formula is as follows:

[0212] Motivition = m i -M f +β

[0213] Ability = a i -A f +α

[0214]

[0215]

[0216] [[ID=2३]]

[0217] Among them,

[0218] Motivition represents the motivation value corresponding to the threshold line;

[0219] Ability represents the ability value corresponding to the threshold line;

[0220] m p represents the predicted value of the motivation value of the worker;

[0221] m i represents the actual motivation value of the i-th worker

[0222] α represents the parameter adjusted on the ability axis;

[0223] β represents the parameter adjusted on the motivation axis;

[0224] lost is the constructed loss function.

[0225] Finally, the values of α and β are obtained.

[0226] Fogg Behavior Model (FBM): The full name is Fogg Behavior Model. The Fogg Behavior Model is an empirical behavior model. Behavior must converge three elements at the same time: motivation, ability, and trigger. When the behavior does not occur, at least one of these three elements is missing. Its horizontal and vertical axes correspond to the levels of motivation and ability respectively, and a threshold line serves as a guide for behavior change.

[0227] S3. Based on the trained worker classification model, select workers with high motivation and high ability from the worker feature dataset to be classified, and recommend tasks to them.

[0228] During specific implementation,

[0229] Based on the trained classification model, identify the highly motivated and highly capable workers in the test set. The specific identification method is as follows:

[0230]

[0231] For example, for the i-th worker, their actual motivation value and ability value are m i , a i respectively. If the following conditions are simultaneously met:

[0232]

[0233]

[0234] Then this worker is identified as a highly motivated and highly capable worker.

[0235] After the identification is completed, calculate the winning bid probability P of the j-th highly motivated and highly capable worker participating in the i-th task ij , and rank the workers for task recommendation according to the winning bid probability and the number of winning bidders allowed for this task to generate the final recommendation list.

[0236]

[0237] Among them,

[0238] P ij represents the winning bid probability value of the j-th worker participating in the i-th task;

[0239] a ij represents the ability value of the j-th worker in the i-th task;

[0240] a ik represents the ability value of the k-th worker in the i-th task;

[0241] W i represents the number of workers participating in the i-th task.

[0242] Embodiment 2

[0243] The present invention also provides a crowdsourcing task personalized recommendation system based on machine learning. The system includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0244] It is understandable that the personalized recommendation system for crowdsourcing tasks based on machine learning provided by the embodiments of the present invention corresponds to the above-mentioned personalized recommendation method for crowdsourcing tasks based on machine learning. For the explanations, examples, beneficial effects, etc. of the relevant content, reference can be made to the corresponding content in the personalized recommendation method for crowdsourcing tasks based on machine learning, which will not be elaborated here.

[0245] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0246] ① Based on the Fogg Behavior Model that integrates motivation and ability theory, the present invention constructs a complete and detailed worker feature recognition system according to the existing literature of crowdsourcing competitions and the available information on crowdsourcing platforms. At the same time, by crawling the historical data of workers and the task data for a period of time, the current worker-task interaction data is constructed to form a worker dataset covering the past and present information of workers, so as to better measure worker features. This helps the platform quickly identify high-value workers, generate a recommendation list based on the selection probability, and finally achieve personalized recommendation of crowdsourcing tasks.

[0247] It should be noted that through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments. In this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to this process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0248] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A personalized recommendation method for crowdsourcing tasks based on machine learning, characterized in that, The method includes: Obtaining historical worker feature data and task feature data; and obtaining worker-task interaction feature data based on the task feature data; Training a worker classification model considering worker motivation and worker ability based on the preprocessed historical worker feature data and worker-task interaction feature data; Based on the trained worker classification model, screening out workers with high motivation and high ability from the worker feature dataset to be classified, and making task recommendations for them; The training of the worker classification model considering worker motivation and worker ability based on the preprocessed historical worker feature data and worker-task interaction feature data includes: S201. Constructing ability indicators and motivation indicators for workers; S202. Determining the weights of each variable of the worker's ability indicators and the weights of each variable of the motivation indicators to obtain the motivation value and ability value of each worker; S203. Obtaining a threshold line model for use as a criterion for dividing workers with high motivation and high ability; The ability indicators of the worker include: inherent ability, professional ability, general experience, professional experience, diversity, complexity; The motivation indicators of the worker include: enjoyment and fun, work autonomy, task complexity, self-marketing / sense of belonging; And the inherent ability includes: platform score, platform certification level, number of past winning bids; The professional ability includes: number of certification categories, number of certification sub-categories, number of certification skills, number of skills over ten years; The general experience includes: number of past task recommendations, number of recently participated tasks, number of recently won tasks, hourly rate; The professional experience includes: number of certified industries, number of tasks included in recently participated tasks, number of tasks included in recently won tasks; The diversity includes: number of recently participated different task categories, number of recently participated different task sub-categories, number of recently won different task categories, number of recently won different task sub-categories; The complexity includes: average number of bidders for recently participated tasks, average number of suggestions for recently participated tasks, average number of favorites for recently participated tasks, average number of views for recently participated tasks; The enjoyment and fun include: average task reward, task approval rate; The work autonomy includes: average recruitment time, task completion rate; The task complexity includes: ratio of average number of favorites to number of bidders for participated tasks, ratio of average number of favorites to number of bidders for won tasks, ratio of average number of views to number of favorites for participated tasks, ratio of average number of views to number of favorites for won tasks; The self-marketing / sense of belonging includes: positive review rate, number of comments; The threshold line model includes: Motivition = m i -M f +β Ability=a i -A f +α Among them, a i represents the magnitude of the ability value of the i-th worker; m i represents the magnitude of the motivation value of the i-th worker; M f Kolmogorov mean value for worker motivation; A f is the Kolmogorov mean of the worker's ability; α represents the parameter adjusted on the ability axis; β represents the parameter adjusted on the motivation axis; And by constructing a loss function lost, training α and β using five-fold cross-validation to make lost converge, and taking the average value of α and β as the final result.

2. The personalized recommendation method for crowdsourcing tasks based on machine learning according to claim 1, wherein, The obtaining of the historical worker feature data and task feature data; And the obtaining of the worker-task interaction feature data based on the task feature data includes: S101. Crawling all historical worker feature data and task feature data within a period of time from the crowdsourcing platform; S102. Preprocess the task feature data and encode the task feature data to obtain a task feature data set; S103. Divide the task feature data set into a training set and a test set in a 1:1 ratio according to the time sequence, and obtain worker-task interaction feature data based on the training set; S104. Preprocess the worker historical feature data and encode it to obtain a worker historical feature data set; S105. Preprocess the worker-task interaction feature data and encode it.

3. The personalized recommendation method for crowdsourcing tasks based on machine learning according to claim 2, characterized in that, The preprocessing of the task feature data includes: Filter out task feature data with the number of task bidders greater than or equal to M1; Exclude tasks where the recruitment has not ended or non-competition tasks; Exclude tasks with missing key features such as the elected reward of the task, the task release time, and the task deadline; Remove duplicates from the repeated task feature data in the list; Exclude outliers in the task list; The preprocessing of the worker historical feature data includes: Exclude workers with the number of past elected suggestions less than or equal to M2; Exclude workers with the number of past suggestions less than or equal to M3; The preprocessing of the worker-task interaction feature data includes: Exclude workers with the number of participated tasks less than M4; Exclude workers with the number of winning tasks less than M5; Complete the missing values.

4. A personalized recommendation method for crowdsourcing tasks based on machine learning according to claim 1, characterized in that The determination of the weights of the variables of the ability index and the weights of the variables of the motivation index of the worker to obtain the motivation value and ability value of each worker includes: Determine the weights of the variables based on the CRITIC objective weighting method; And the calculation formulas for the motivation value and ability value of the worker are as follows: a i represents the magnitude of the ability value of the i-th worker; m i represents the magnitude of the motivation value of the i-th worker; Indicates the objective weight of the j-th ability variable; Represents the magnitude of the objective weight of the k-th motivation variable; x ij represents the value of the j-th ability variable of the i-th worker; y ik represents the value of the k-th motivation variable of the i-th worker.

5. A personalized recommendation method for crowdsourcing tasks based on machine learning according to claim 1, characterized in that, Based on the trained worker classification model, screening out workers with high motivation and high ability from the to-be-classified worker feature data set includes: when the motivation value and ability value of the to-be-classified worker are both greater than the threshold line, they are identified as workers with high motivation and high ability.

6. A crowdsourcing task personalized recommendation system based on machine learning, the system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1-5 above.

Images