CN109711653B - Weike task recommendation method based on Weike-task-label three-square diagram - Google Patents
Weike task recommendation method based on Weike-task-label three-square diagram Download PDFInfo
- Publication number
- CN109711653B CN109711653B CN201810034199.8A CN201810034199A CN109711653B CN 109711653 B CN109711653 B CN 109711653B CN 201810034199 A CN201810034199 A CN 201810034199A CN 109711653 B CN109711653 B CN 109711653B
- Authority
- CN
- China
- Prior art keywords
- task
- resources
- tasks
- label
- weike
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Abstract
The invention provides a recommendation method for a Weike task, which combines historical information of the Weike with a task label by adopting a Weike-task-label trimap model so as to better match the Weike with the task, and comprises the following steps: (1) initializing a Weike-task-label three-square diagram according to historical interaction data of the Weike and the tasks; (2) performing primary resource diffusion and transfer on the Wickey-task bipartite graph, distributing resources on the tasks to the Wickeys, and then re-distributing the resources back to the tasks; meanwhile, the diffusion and the transfer of the resources are carried out on the task-label bipartite graph once, the resources on the task are distributed to the labels and then are redistributed back to the task; (3) and integrating the resources redistributed to the tasks on the Wieker-task bipartite graph and the task-label bipartite graph according to a certain weight, and recommending the tasks according to the distribution of the integrated resources from big to small.
Description
Technical Field
The invention relates to the field of personalized recommendation, in particular to a task recommendation method for a Wieker website.
Background
With the rapid development of information communication technology and internet technology, we gradually enter the big data age. People are surrounded by mass data generated by information explosion, so that the problem of information overload is caused, and the use efficiency of information is reduced. The personalized recommendation system becomes a very potential solution to the information overload problem, and finds the potential interest of the user by analyzing the interest characteristics of the user, the attribute characteristics of articles, the historical behavior record of the user and other information, predicts the preference of the user, and recommends the information more interested by the user to the corresponding user, thereby achieving the effect of personalized recommendation.
Since the beginning of the century, the internet began to develop rapidly, various innovative applications and new concepts of the internet continuously appeared, and how to use the internet for knowledge management has attracted high attention of many scholars in the internet world and the knowledge management world. The wecker mode is generated in the big background, and is a network innovation mode for knowledge management by using the internet. The wechat mainly refers to a group of people who give knowledge, skills, expertise, experience and the like to others through a network platform and exchange the knowledge, the skills, the expertise, the experience and the like for actual economic benefits. The appearance of the Wickey mode brings great convenience to our life, not only can solve the problem that a search engine cannot creatively give answers and improve the enthusiasm of users for using the Internet, but also can provide a flexible employment mode and enrich the types of Internet services.
However, the wippen model has some defects, the employer and the wippen do not know each other, and the information is often mismatched, so that the employer cannot find the proper wippen to help him to solve the problems, the wippen cannot find the proper task to complete, and a recommendation system based on the wippen can solve the problems. The wecker based recommendation system provides the employer with the best available weckers to complete their mission, and matches the best available weckers to the mission they are solving, thereby minimizing the distance between the employer and the weckers.
The label is a means for labeling information, and Vig et al predicts the User' S preference for the label by analyzing the labeling behavior of the User and the scoring behavior of the film and label on the MovieLens film label dataset (see Vig J, Sen S, rice J. tags: displaying and recording uses [ C ]// International Conference on Intelligent User interfaces. acm,2009: 47-56.). The label-based recommendation algorithm is easy to understand and implement, but is not systematic and theoretical enough. The recommendation algorithm based on the graph makes up the defects in the aspect, and the recommendation becomes mathematical and rigorous through the combination of the label and the graph. Zhou et al introduced personalized recommendation problems into a bipartite network, where the user served as one class of nodes in the bipartite graph, the item served as another class of nodes in the bipartite graph, and the user's actions such as evaluation or operation on the item served as edges between the nodes. The correlation weight can be calculated between the same type of nodes through another type of nodes, and the calculation mode adopts a resource allocation mode based on a binary network (see Zhou T, Ren J, Medo M, Zhang Y-C (2007) duplicate network project and personal communication. Phys Rev E76: 046115 for details).
Disclosure of Invention
Based on the background technology, the invention provides a task recommendation method for a Wiekang website, which adopts a model of a Wiekang-task-label three-division graph, fully utilizes various data information, and combines historical information of the Wiekang and the task with a label of the task, so as to carry out comprehensive personalized recommendation for the Wiekang, and better match between the Wiekang and the task can be realized, and the method comprises the following steps:
(1) initializing a Weike-task-label three-square diagram according to historical interaction data of the Weike and the tasks;
(2) performing primary resource diffusion and transfer on the Wickey-task bipartite graph, distributing resources on the tasks to the Wickeys, and then re-distributing the resources back to the tasks; meanwhile, the diffusion and the transfer of the resources are carried out on the task-label bipartite graph once, the resources on the task are distributed to the labels and then are redistributed back to the task;
(3) and integrating the resources redistributed to the tasks on the Wieker-task bipartite graph and the task-label bipartite graph according to a certain weight, and recommending the tasks according to the distribution of the integrated resources from big to small.
Drawings
FIG. 1 is a schematic diagram of the recommendation of the method of the present invention
A schematic diagram of a three-dimensional chart containing 3 guests, 5 tasks and 4 tags. FIG. (a) shows the initial state of the ternary diagram when the target Weker is set to U1; FIG. (b) depicts the result after the first step of diffusion, where the resources on the task have been transferred to the Wigner and the tag; finally, the final result after two diffusions is shown in the diagram (c), and the resource is transferred to the task again.
FIG. 2 shows a graph of the experimental results of the method of the present invention.
Detailed Description
A personalized task recommendation method for a Wiekang website based on a Wiekang-task-label three-square diagram comprises the following steps: firstly, initializing a Weike-task-label three-square diagram according to historical interaction data of Weike and tasks; then, performing primary resource diffusion and transfer on the Wickey-task bipartite graph, distributing resources on the tasks to the Wickeys, and then redistributing the resources back to the tasks; meanwhile, the diffusion and the transfer of the resources are carried out on the task-label bipartite graph once, the resources on the task are distributed to the labels and then are redistributed back to the task; and finally, integrating the resources redistributed to the tasks on the Wieker-task bipartite graph and the task-label bipartite graph according to a certain weight, and recommending the tasks according to the integrated resource distribution in a descending order. The detailed process is shown in FIG. 1.
The key point of the implementation of the invention is 3 points: resource initialization of a Wicker-task-label trigonometric diagram, resource transfer on the Wicker-task bipartite diagram and the task-label bipartite diagram, and integration of reallocated resources on tasks according to certain weight. The main implementation details are presented below:
1. resource initialization for Wicker-task-tag trimap
Assuming that a resource exists on each task initially, the more resources the more attractive the task is to the wippen, the more worthwhile the task is to be recommended to the wippen.
Give a Wike UiHis initial resource vector can be set
Comprises the following steps:
fj(Ui)＝aij j 1,2, …, m formula (1)
At this time
Is a vector of length m consisting of 0 and 1, representing the initial resource distribution for each task, i.e. if wecker UiAfter the task is completed, the initial resource on the task is set to be 1, otherwise, the initial resource on the task is set to be 0. Wherein different initial resource distributions for different customers represent personalized preferences for different customers.
2. Resource migration on Wicker-task bipartite graph and task-tag bipartite graph
For the wiki-task bipartite graph, a task will evenly distribute this resource to each wiki that is adjacent to it (the wiki is considered to be adjacent if it has completed the task). Thus, through the first diffusion, resources are transferred from the task to the wippen. Specifically, the number of guests adjacent to each task is obtained, and then the size of resources which are evenly distributed to each guest by the tasks is calculated; for each wife, the sum of the evenly allocated resources across all of his neighboring tasks is summarized. At this point the first diffusion is complete and the resource has been transferred from the task to the guest. Then the visitors redistribute the received resources to the tasks adjacent to the visitors evenly through the second diffusion. Similar to the allocation of resources from tasks to guests, the number of tasks adjacent to each guest is firstly obtained, and then the size of the resources evenly allocated to each task is calculated; and for each task, summarizing the sum of the evenly distributed resources on all the customers adjacent to the task to obtain the final task resource distribution. Task resource distribution
The specific calculation formula is as follows:
wherein the content of the first and second substances,
indicates the number of adjacent tasks of Wiskl, fsRepresenting the initial number of resources on the task s,
representing the number of adjacent victims of task s.
For a task-label bipartite graph, a task will evenly allocate this resource to each of its adjacent labels (if the task includes the label, they are considered to be adjacent). Thus, through the first diffusion, resources are transferred from the task to the targetAnd (6) labeling. The tags will redistribute their received resources equally to their neighboring tasks through a second diffusion. And sequencing the tasks according to the size of the resources on the tasks after the two times of diffusion, and recommending the first N tasks with the largest resource values to the user. After two diffusions, the final task resource distribution
Can be expressed as:
wherein the content of the first and second substances,
number of adjacent tasks, f, representing label lsRepresenting the initial number of resources on the task s,
indicating the number of adjacent tags for task s. And circularly processing all the users to obtain recommendation results of all the users.
3. Integrating the re-distributed resources on the task according to a certain weight
A parameter alpha is set to represent the weight distribution of the Weike-task bipartite graph and the task-label bipartite graph on the assigned task resources,
can be expressed as:
where α is a parameter that can be set, in the extreme case α ═ 1 and α ═ 0 represent the calculation by the wikswage-task bipartite algorithm alone or by the task-label bipartite algorithm alone, respectively. The parameter can be adjusted according to the actual situation, so that the final parameters such as accuracy, recall rate and the like reach a relatively optimal result.
And (3) experimental verification:
we use the trading data of 1 month of 2014 by wecker. The data provided by the wecker company can be divided into 9 tables, namely a personal information table, a task manuscript table, a task table, an integrity guard table, an industry/skill table, a user skill recording table, a user expansion table, a user certificate expansion table and a collection table. The experimental effect is mainly reflected by the recall ratio and the accuracy, and fig. 2 is a graph of the recall ratio and the accuracy varying with the weight alpha. Where the length of the recommendation list for the experiment is set to 10, α can be understood as the proportion of resource allocation on the mission, and if α is greater than 0.5, it is more inclined to score the resources according to the wecker-mission bipartite graph, i.e. to allocate more resources to the wecker, whereas if α is less than 0.5, it is more inclined to the mission-label bipartite graph, i.e. to allocate more resources to the labels. Here, α starts at 0, increments by 0.1 each time, and ends at 1. When alpha is equal to 0, the calculation is equivalent to only adopting the task-label bipartite graph, the accuracy is 0.0159, the recall rate is 0.0554, and the F value is 0.0247; when alpha is 1, the calculation is equivalent to only adopting a Wicker-mission bipartite graph, the accuracy rate is 0.0661, the recall rate is 0.2308, and the F value is 0.1027; when α is 0.8, the accuracy becomes 0.0665, the recall becomes 0.2323, and the F value becomes 0.1034, which is better than the effect obtained by using only the wiki-task bipartite graph or the task-label bipartite graph, so that 0.8 is the optimal α value.
Claims (1)
1. A personalized task recommendation method for a Wiekang website based on a Wiekang-task-label three-square diagram comprises the following steps:
(1) initializing a Weike-task-label three-square diagram according to historical interaction data of the Weike and the tasks;
(2) performing primary resource diffusion and transfer on the Wickey-task bipartite graph, distributing resources on the tasks to the Wickeys, and then re-distributing the resources back to the tasks; meanwhile, the diffusion and the transfer of the resources are carried out on the task-label bipartite graph once, the resources on the task are distributed to the labels and then are redistributed back to the task;
(3) integrating resources redistributed to tasks on the Wieker-task bipartite graph and the task-label bipartite graph according to a certain weight, and recommending the tasks according to the sequence from big to small according to the integrated resource distribution;
initializing a Weike-task-label trigonometric chart according to historical interaction data of the Weike and the task in the step (1); specifically, assuming that a resource exists on each task in the initial situation, the more the resource, the more the task has the greater attraction to the wife, the more the task is worth being recommended to the wife; give a Wike UiSetting up its initial resource vector
Comprises the following steps:
fj(Ui)＝aij，j＝1，2，…，m
at this time
Is a vector of length m consisting of 0 and 1, representing the initial resource distribution for each task, i.e. if wecker UiAfter the task is completed, setting the initial resource on the task as 1, otherwise, setting the initial resource on the task as 0; wherein different initial resource distributions for different tenants represent personalized preferences for different tenants;
in the step (2), resource transfer is carried out on the Wicker-task bipartite graph and the task-label bipartite graph; for the Weike-task bipartite graph, the task evenly distributes this resource to each Weike adjacent to it, and if the Weike completes the task, they are considered adjacent; thus, through the first diffusion, resources are transferred from tasks to guests; specifically, the number of guests adjacent to each task is obtained, and then the size of resources which are evenly distributed to each guest by the tasks is calculated; for each wife, sum up the evenly allocated resources across all of his adjacent tasks; at this point the first diffusion is complete and the resource has been transferred from the task to the guest; then, theThe weckers evenly redistribute the resources they receive to the tasks adjacent to them through the second diffusion; similar to the allocation of resources from tasks to guests, the number of tasks adjacent to each guest is firstly obtained, and then the size of the resources evenly allocated to each task is calculated; for each task, summarizing the sum of the evenly distributed resources on all the guests adjacent to the task to obtain the final task resource distribution; task resource distribution
The specific calculation formula is as follows:
wherein the content of the first and second substances,
indicates the number of adjacent tasks of Wiskl, fsRepresenting the initial number of resources on the task s,
represents the number of adjacent victims of task s;
for a task-label bipartite graph, a task evenly distributes resources to each label adjacent to it, and if the task includes the label, the tasks are considered adjacent; thus, through the first diffusion, the resource is transferred from the task to the label; the tags redistribute the resources they receive evenly to their adjacent tasks through a second diffusion; sorting the tasks according to the size of the resources on the tasks after the two-time diffusion, and recommending the first N tasks with the largest resource values to the user; after two diffusions, the final task resource distribution
Expressed as:
wherein the content of the first and second substances,
number of adjacent tasks, f, representing label lsRepresenting the initial number of resources on the task s,
the number of adjacent tags representing task s; performing cyclic processing on all users to obtain recommendation results of all users;
in the step (3), the resources redistributed on the task are integrated according to a certain weight; specifically, a parameter alpha is set to represent the weight distribution of the Weike-task bipartite graph and the task-label bipartite graph on the task resources,
expressed as:
α is a control variable, and in the extreme case α ═ 1 and α ═ 0 represent the calculation by the wecker-task bipartite algorithm alone or by the task-label bipartite algorithm alone, respectively.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711020996 | 2017-10-26 | ||
| CN2017110209962 | 2017-10-26 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109711653A CN109711653A (en) | 2019-05-03 |
| CN109711653B true CN109711653B (en) | 2020-12-15 |
Family
ID=66253626
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810034199.8A Active CN109711653B (en) | 2017-10-26 | 2018-01-15 | Weike task recommendation method based on Weike-task-label three-square diagram |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109711653B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111400483B (en) * | 2020-03-17 | 2022-06-21 | 重庆邮电大学 | Time-weighting-based three-part graph news recommendation method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559626A (en) * | 2013-09-24 | 2014-02-05 | 浙江工商大学 | Individualized commodity recommendation method based on bigraph resource non-uniform distribution |
| CN104899763A (en) * | 2015-05-07 | 2015-09-09 | 西安电子科技大学 | Personalized recommendation method based on bilateral diffusion of bipartite network |
| CN106934252A (en) * | 2017-03-08 | 2017-07-07 | 华南理工大学 | A kind of triple net Resources Spread method |
-
2018
- 2018-01-15 CN CN201810034199.8A patent/CN109711653B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559626A (en) * | 2013-09-24 | 2014-02-05 | 浙江工商大学 | Individualized commodity recommendation method based on bigraph resource non-uniform distribution |
| CN104899763A (en) * | 2015-05-07 | 2015-09-09 | 西安电子科技大学 | Personalized recommendation method based on bilateral diffusion of bipartite network |
| CN106934252A (en) * | 2017-03-08 | 2017-07-07 | 华南理工大学 | A kind of triple net Resources Spread method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109711653A (en) | 2019-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10841743B2 (en) | Branching mobile-device to system-namespace identifier mappings | |
| CA2619076C (en) | Scalable user clustering based on set similarity | |
| AU2014201827B2 (en) | Scoring concept terms using a deep network | |
| Brah et al. | Heuristics for scheduling in a flow shop with multiple processors | |
| CN110147882B (en) | Neural network model training method, crowd diffusion method, device and equipment | |
| CN107969156A (en) | For handling the neutral net of graph data | |
| CN102591915A (en) | Recommending method based on label migration learning | |
| CN107247753B (en) | A kind of similar users choosing method and device | |
| CN106651544A (en) | Conversational recommendation system for minimum user interaction | |
| Kotzias et al. | Home is where your friends are: Utilizing the social graph to locate twitter users in a city | |
| CN109410001A (en) | A kind of Method of Commodity Recommendation, system, electronic equipment and storage medium | |
| CN107480227A (en) | A kind of information-pushing method based on big data | |
| JP6308339B1 (en) | Clustering system, method and program, and recommendation system | |
| CN109711653B (en) | Weike task recommendation method based on Weike-task-label three-square diagram | |
| CN110458602A (en) | Method of Commodity Recommendation, device, electronic equipment and storage medium | |
| CN109075987A (en) | Optimize digital assembly analysis system | |
| CN113704620A (en) | User label updating method, device, equipment and medium based on artificial intelligence | |
| CN108628990A (en) | Recommendation method, computer installation and readable storage medium storing program for executing | |
| CN110020918B (en) | Recommendation information generation method and system | |
| Souchkov | Systematic business innovation: a roadmap | |
| Pasta et al. | Generating online social networks based on socio-demographic attributes | |
| Hutchinson et al. | Enumeration of octagonal tilings | |
| Kudori | Event recommendation system using hybrid method based on mobile device | |
| KR101985603B1 (en) | Recommendation method based on tripartite graph | |
| CN109388738B (en) | Information pushing method and device |
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 |