Task Assignment of the Improved Contract Net Protocol under a Multi-Agent System
Round 1
Reviewer 1 Report
Authors of the work “Task Assignment of Improving Contact Net Protocol under Multi-Agent System” propose in this paper a method to slove the overall efficiency of existing contract net protocols in the tender issuance stage, and generate a lot of redundant information in the negotiation process.
Overall, it is a well-structured paper; the introduction section is wide and presents the purpose of the research in detail. There is a comparison and evaluations of the proposed method, using proper figures and showing up the proofs of the experiments included in the paper.
Although the proposal is interesting and within the scope of the Algorithms, there are different issues that should be addressed in order to improve the work.
[Minor comments]
· General: Updating with more recent references (Within the last 5 years).
· Improve the state of the art. It is advisable to introduce the concept of agent and multi-agent systems. The first work that shows a broad state of the art about agents and multi-agent systems as a concept of architecture is as follows (There is a part focused on energy optimization that part is not relevant to your work). The other works could help you to improve the agent state of art in order to explain why the use of agents is appropriate and recommended in this work.:
o González-Briones, A., De La Prieta, F., Mohamad, M., Omatu, S., & Corchado, J. (2018). Multi-agent systems applications in energy optimization problems: A state-of-the-art review. Energies, 11(8), 1928.
o Carbó, J., Molina, J. M., & Patricio, M. A. (2016). Asset Management System through the design of a Jadex Agent System. ADCAIJ: Advances in Distributed Computing and Artificial Intelligence Journal, 5(2), 1-14.
o González-Briones, A., Prieto, J., De La Prieta, F., Herrera-Viedma, E., & Corchado, J. M. (2018). Energy optimization using a case-based reasoning strategy. Sensors, 18(3), 865.
[Major comments]
· The results shown in the Experimental Process and Results section would be scientifically supported if any statistical tests were applied. You still need to report (i) sample size, (ii) alpha value, and clearly state the null and alternative hypotheses. In order to improve the comparative, It would be interesting to look how this is done in the following paper, 4.1 section:
o González-Briones, A., Villarrubia, G., De Paz, J. F., & Corchado, J. M. (2018). A multi-agent system for the classification of gender and age from images. Computer Vision and Image Understanding.
The used English is correct.
Author Response
Dear Editor and Reviewer:
Thank you for your letter and for the professor’s comments concerning our manuscript entitled “Task Assignment of Improving Contract Net Protocol under Multi-Agent System” (ID: 439235). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The main corrections in the paper and the responds to your comments are as flowing:
Point 1: Updating with more recent references (Within the last 5 years).
Response 1: Added references for the last five years, including 4 articles mentioned in the external review comments.
Point 2: The results shown in the Experimental Process and Results section would be scientifically supported if any statistical tests were applied. You still need to report (i) sample size, (ii) alpha value, and clearly state the null and alternative hypotheses.
Response 2: In the first experiment, the rank sum test is used, and the overall sample is set to 24, divided into four groups according to L=1, 2, 3, and 4, each group is the traffic required to complete the unit task. Using the Kruskal-Wallis test, to begin, we established an the test level equal to 0.05 and we proposed a null hypothesis. It is assumed that H0 is the same for the four groups of data, and H1- the overall distribution of each group is not the same.All the data are combined and sorted and given the sequence number value, and the four groups of sequence numbers are respectively summed to obtain R1, R2, R3, and R4, and the H test statistic is calculated again. With the calculation of spss, p=0.38<0.05 is obtained, and H0 is rejected. When the load value is different, the traffic is also different, and when L=3, the overall traffic is the smallest. For the second experiment, the same method was used for statistical test.
Because the sample size is small, the statistical test of the experimental results is simply a simple analysis using software. Expanding the sample size for detailed statistical testing is the focus of the next step.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper is technically well developed and relevant to this journal as it comes to a task assignment method, with a rich field of research which has been increasingly addressed in recent years. The method is an improvement of the Contract Net Protocol. The improved contract net protocol can reduce the network communication and run-time compared with classical contract net and dynamic contract net protocol, and the algorithm has high task completion rate and multi-agent coordination ability. Therefore, this work presents significant contributions.
I thought part of the description of some points was missing. I came away with some questions. Therefore, I recommend that a minor revision is warranted. I explain my concerns in more detail below. I ask the authors to specifically address each of my comments in their response.
1. In formula 1, the explanation of the variables could be more explicable, describing the meaning of Y, λ1 and λ2.
2. In figure 3, specifically in the decision process "Whether the second in the sequence table gets the task" , are the “Yes” or “No” decisions reversed?
3. In table 2, It has been explained that more errors occur in the other algorithms. In addition, the proposed algorithm can handle failures and complete failed tasks. But it could be clearer why more errors occur in the other algorithms. It seems that it is due to the choice of the agent that has the greatest ability to perform the task, but this is not explicit.
4. The word "contract" is incorrectly written in various parts of the text ("contact" instead of "contract"), including the title.
5. The experiments evaluated the run-time and the network traffic and found that the proposed method is more advantageous at these points. However, the critical points of the new method were not mentioned. For example, the impact on the storage capacity of the agents, once they record the processes.
6. In line 147, check space size at the beginning of the line.
7. In line 163, the formula number is wrong. It's 5, not 3.
8. In line 170, the formula number is wrong. It's 6, not 4.
9. In line 227, check the grammar “… is are …”
Author Response
Dear Editor and Reviewer:
Thank you for your letter and for the professor’s comments concerning our manuscript entitled “Task Assignment of Improving Contract Net Protocol under Multi-Agent System” (ID: 439235). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The main corrections in the paper and the responds to your comments are as flowing:
Point 1: In formula 1, the explanation of the variables could be more explicable, describing the meaning of Y, λ1 and λ2.
Response 1: The increase and decrease of the value are nonlinear, and the more successfully completed tasks, the greater the increase of the value; at the same time, the more the task fails, the greater the value decreases. Where Y represents the magnitude of the increase in value, the agent that successfully completes multiple tasks in succession is larger than the Y value of the agent that successfully completes the task, and is adjusted by using and .
Point 2: In figure 3, specifically in the decision process "Whether the second in the sequence table gets the task" , are the “Yes” or “No” decisions reversed?
Response 2: Without reversing, when the second agent of the decision table does not get the task, it is in an idle state, and can directly give the task to it. When the agent already has the task, the manager agent must re-sensing the bid pheromone from the environment.
Point 3: In table 2, It has been explained that more errors occur in the other algorithms. In addition, the proposed algorithm can handle failures and complete failed tasks. But it could be clearer why more errors occur in the other algorithms. It seems that it is due to the choice of the agent that has the greatest ability to perform the task, but this is not explicit.
Response 3: It has been explained in Section 5.3 of the manuscript. Because the traditional contract net protocol does not respond to the threshold and trust degree, the number of task errors is relatively large, and the dynamic contract net protocol has trust degree, but it is only linearly increased and reduced, and it is impossible to screen out the contractor agent with real ability.
Point 4: The word "contract" is incorrectly written in various parts of the text ("contact" instead of "contract"), including the title.
Response 4: The ‘contact’ in the text has been completely replaced by ‘contract’.
Point 5: The experiments evaluated the run-time and the network traffic and found that the proposed method is more advantageous at these points. However, the critical points of the new method were not mentioned. For example, the impact on the storage capacity of the agents, once they record the processes.
Response 5: The conclusion part of the manuscript adds a brief overview of the shortcomings of improving the contract net protocol. In the whole process, the manager agent and the contractor agent only update and maintain the pheromone parameter calculation and decision table, occupying a large storage capacity in the public environment.
Point 6: In line 147, check space size at the beginning of the line.
Response 6: Indented the beginning of line 147, and modified at line 156.
Point 7: In line 163, the formula number is wrong. It's 5, not 3.
Response 7: Indented the beginning of line 163, and modified at line 173.
Point 8: In line 170, the formula number is wrong. It's 6, not 4.
Response 8: Indented the beginning of line 170, and modified at line 180.
Point 9: In line 227, check the grammar “… is are …”
Response 9: In line 227 has deleted ‘are’, and modified at line 246.
Author Response File: Author Response.pdf
Reviewer 3 Report
In this paper authors propose an improvement of the Contact Net Protocol. Unfortunately I am unable to review the paper because the authors do not provide a definition of the problem they are addressing, therefore it is impossible to state the correctness of the proposed protocol, and also to judge its improvements (e.g., which is the measure of efficiency?). For completeness, and also to help the reader that is not so familiar with the specific protocol, they should also provide a high level description of the protocol they are improving.
Abstract:
- the sentence in "Background" does not seem sound
- what is L_max? the reader has no idea at this point in the paper
Introduction:
at the end, when authors state their results, they should specify in which section such results are shown.
Author Response
Dear Editor and Reviewer:
Thank you for your letter and for the professor’s comments concerning our manuscript entitled “Task Assignment of Improving Contract Net Protocol under Multi-Agent System” (ID: 439235). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The main corrections in the paper and the responds to your comments are as flowing:
Point 1: Unfortunately I am unable to review the paper because the authors do not provide a definition of the problem they are addressing, therefore it is impossible to state the correctness of the proposed protocol, and also to judge its improvements (e.g., which is the measure of efficiency?). For completeness, and also to help the reader that is not so familiar with the specific protocol, they should also provide a high level description of the protocol they are improving.
Response 1: The contract net protocol is a classic method to solve the task assignment problem. The purpose of this paper is to solve the two problems existing in the existing contract net protocol, mainly the improvement of the existing algorithm. In the experimental part of the article, multiple agent models were established to assign tasks, and the advantages of the improved algorithm were verified. Therefore, I don't think it is necessary to use a specific task assignment problem to verify the new algorithm.
Because it is a task-sharing method under distributed systems, the current contract net protocol uses traffic and run-time to measure the pros and cons of the algorithm. This paper also explains the advantages of the new algorithm based on the allocation results and task completion rates of the three algorithms. Section 3.3 of the paper uses a flow chart to describe the algorithm in a high-level way. Don't know if this is the case?
Point 2: The sentence in "Background" does not seem sound
Response 2: The sentence in "Background" has been modified.
Point 3: What is L_max? the reader has no idea at this point in the paper
Response 3: L_max is the maximum load value and has been briefly explained in Results.
Point 4: At the end, when authors state their results, they should specify in which section such results are shown.
Response 4: The results of the three experiments in 5.1, 5.2, and 5.3 of the fifth section of the paper are explained, and has been added in Introduction
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The article has been substantially improved. It is recommended that it be accepted in its present form for publication.
Author Response
Your opinion has helped me a lot, and we are feel so warm for your suggestions.
Reviewer 3 Report
Authors propose a protocol to improve the efficiency of a task assignment problem under the multi agent system. Reported results actually show that the protocol generates less traffic in the system as is faster than competitor protocols. However, the paper is unpleasant to read (due to lack of formalism and accuracy, grammar errors and weird sentence constructions), too many questions arise while reading it and the answers are not always obvious to find or has to be guessed. Therefore it is difficult to assess overall soundness of the proposed protocol.
I still think that many answers would come naturally to the reader if a formal definition of the problem is given at the beginning of the paper in a separate section between the introduction and what is now section 2. I means something like:
we are given a MAS with n agents connected by a ... (specify network topology) .Each agent is characterized by a capabilty that represents its ... (whatever it is) and might vary over time according to .... and then list all other values that characterize an agent. We are given m tasks: each task k is characterized by an arrival time? a priority? a deadlne? each task k can be decomposed into m_k subtask (each with a completion time?) and specify if subtasks inherit the task priority or it if it has its priority that might be different for the other subtasks of the same task... and so on...
Therefore I do not suggest that the paper is accepted in this present form.
Some examples of lack of formalism and accuracy (and related arising questions):
- It is not given to know if the network topology connecting the agents is known or not, and if this is important for the problem (or any network would do the job)
- pag 2, line 1 e 2: where are time and deadlines taken into consideration in the proposed protocol and in the experiments?
- Pag 4: what is the difference between a "completed task" and a "successfully completed task"?
- pag 4: what kind of "adjustment" is introduced in formula (1) by means of \lambda_1 e \lambda_2? some kind of explanation should be given.
- pag 4: is task difficulty directly proportional to its priority (or viceversa)? or might it happen that a task has high priority even if it is easy? what happens to \alpha and \beta in this case?
- pag 5: is the way in which tasks are decoupled given as an input of the problem?
- pag 5: ID and Nam seems to be two identifiers for the same thing, the task. What is the point in having two identifiers? is Nam the identifier of the sub task? and Pri, is it the original task priority or the subtask one?
- pag 5 sec 3.2: what is the agent capability? how does an agent estimate CostT: does each task come with an indication on completion time?
- pag 5, eq (3): it seems to me that \alpha and \beta are independent from the contractor, but depend only on the task. Why is their calculation left to contractor? could they just replace Pri and Rew in pheromone A? is there any sense in the definition of \alpha and \beta?
-pag 5, eq (4): what is r? why the 0 in T_0_i?
- pag 6, line + 4: when does the manager decide that it has all the bidding? might this policy influence running time and congestion?
- pag 6, line + 4: "In the case of the same task, the priority of the task is sorted": same ID or same Nam? and why should the priority of the same task be different form bid to bid?
- pag 6 eq (5): any intuition behind the formula? and what about \lambda_2 and \lambda_3?
-pag 5 step5: how does the manager inform the winning contractor? there is no contractor ID in the bidding pheromone.
-pag 6 step 6, last sentence is not clear: who is the first agent and who is the second one?
- In section 4.2 it seems that the MAS used for the experiments has 10 agents, but in section 5 write "when the number of agents is fixed at 20..."
- In section 4.2 authors introduce the concept of "task set", but previously they only talked about a task being decoupled. Is a task set the outcome of decoupling a task?
- In section 5.3 authors talk about "errors", do they mean tasks that have been abandoned by an agent?
Some examples of grammar errors and weird sentences:
- First sentence of the abstract.
- From introduction: "the ability of the agent to change during the execution of the mission may change" what does the first "change" refer to? change what?
- pag 4: "the magnitude of \alpha is proportional to it", what does it stand for? "the difficulty and the priority" is plural.
- end of pag 4: "while the improved ... protocol in this study reduces mxn communication times"
- pag 5: "Rew represents used", "Cap indicated as"
- check "s" for plural and third person
Author Response
Dear Editor and Reviewer:
Thank you for your letter and for the professor’s comments concerning our manuscript entitled “Task Assignment of Improving Contract Net Protocol under Multi-Agent System” (ID: 439235). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The main corrections in the paper and the responds to your comments are as flowing:
Point 1: We are given a MAS with n agents connected by a ... (specify network topology) .Each agent is characterized by a capabilty that represents its ... (whatever it is) and might vary over time according to .... and then list all other values that characterize an agent. We are given m tasks: each task k is characterized by an arrival time? a priority? a deadlne? each task k can be decomposed into m_k subtask (each with a completion time?) and specify if subtasks inherit the task priority or it if it has its priority that might be different for the other subtasks of the same task... and so on...
Response 1: I have already added relevant content before section 2.1.
Point 2: It is not given to know if the network topology connecting the agents is known or not, and if this is important for the problem (or any network would do the job)
Response 2: The contract net protocol is a typical method to solve the task assignment under the distributed structure. Therefore, it is generally considered to adopt a distributed topology structure and is described in the text.
Point 3: pag 2, line 1 e 2: where are time and deadlines taken into consideration in the proposed protocol and in the experiments?
Response 3: I have added a deadline to the contract pheromone.
Point 4:- Pag 4: what is the difference between a "completed task" and a "successfully completed task"?
Response 4: Completed task means that the negotiation is successful, the task is assigned to the contractor agent to complete, and part of the task is abandoned in the middle. Successfully completed task means the completion of the task
Point 5:- pag 4: what kind of "adjustment" is introduced in formula (1) by means of \lambda_1 e \lambda_2? some kind of explanation should be given.
Response 5: The value of Y is adjusted by Lambda 1 and Lambda 2, which in turn affects the response threshold. When the number of times the bid is obtained C is small, the lambda 2 is larger; otherwise, the lambda 1 is larger, and as C increases, the lambda 1 and the lambda 2 are linearly changed.
Point 6:- pag 4: is task difficulty directly proportional to its priority (or viceversa)? or might it happen that a task has high priority even if it is easy? what happens to \alpha and \beta in this case?
Response 6: The priority of the task is given in advance, and there is no special correspondence between the remaining difficulties. It can also be a very simple task, but it has a shorter deadline, so it has a higher priority.
Point 7:- pag 5: is the way in which tasks are decoupled given as an input of the problem?
Response 7: The decoupling of tasks is beyond the scope of this paper, mainly to distribute the decoupled tasks. Of course, we use a flexible time task tree for decoupling according to the basic process of operations.
Point 8:- pag 5: ID and Nam seems to be two identifiers for the same thing, the task. What is the point in having two identifiers? is Nam the identifier of the sub task? and Pri, is it the original task priority or the subtask one?
Response 8: Nam and ID are identifiers of the same task. Nam mainly helps in the research of the situation display, so it is not deleted in this article. Pri is the priority of the subtask
Point 9:- pag 5 sec 3.2: what is the agent capability? how does an agent estimate CostT: does each task come with an indication on completion time?
Response 9: The function of the contractor agent is to calculate according to the task pheromone. Each type of task will have a rough completion time. The contractor agent will finally estimate the completion time according to its own capabilities, plus the rest of the tasks to be completed are the CosT.
Point 10:- pag 5, eq (3): it seems to me that \alpha and \beta are independent from the contractor, but depend only on the task. Why is their calculation left to contractor? could they just replace Pri and Rew in pheromone A? is there any sense in the definition of \alpha and \beta?
Response 10: The alpha and beta values are used to determine the response threshold change, so its calculations should be left at the end and determined by the manager agent based on the task completion. Because Pri and Rew are parameters used in the task bidding process, they cannot be replaced.
Point 11:-pag 5, eq (4): what is r? why the 0 in T_0_i?
Response 11: r is the number of tasks to be completed by the contractor agent, and 0 is to distinguish multiple agents.
Point 12:- pag 6, line + 4: when does the manager decide that it has all the bidding? might this policy influence running time and congestion?
Response 12: The manager agent has set a time. When the time is up, the bidding must be made. The relevant instructions have been added in the manuscript.
Point 13:- pag 6, line + 4: "In the case of the same task, the priority of the task is sorted": same ID or same Nam? and why should the priority of the same task be different form bid to bid?
Response 13: This sentence is wrong, it should be “In the case of the same task set”.
Point 14:- pag 6 eq (5): any intuition behind the formula? and what about \lambda_2 and \lambda_3?
Response 14: I don't know if you are talking about Lambda 3 and Lambda 4. These two parameters are calculated by fitting in the scene we set.
Point 15:-pag 5 step5: how does the manager inform the winning contractor? there is no contractor ID in the bidding pheromone.
Response 15: H is added to the bid pheromone to represent the ID number of the contractor agent.
Point 16:-pag 6 step 6, last sentence is not clear: who is the first agent and who is the second one?
Response 16: In step3, the task and evaluation sequence table are obtained, so when the first contractor agent cannot complete the task, the second agent in the table is selected to complete.
Point 17:- In section 4.2 it seems that the MAS used for the experiments has 10 agents, but in section 5 write "when the number of agents is fixed at 20..."
Response 17: The 10 agents set up in the initial experiment, and later re-set the experiment, but forgot to change it in the text, now changes have been made.
Point 18: In section 4.2 authors introduce the concept of "task set", but previously they only talked about a task being decoupled. Is a task set the outcome of decoupling a task?
Response 18: Multiple tasks are decoupled into multiple subtasks, and multiple subtasks of the same type form a task set.
Point 19:- In section 5.3 authors talk about "errors", do they mean tasks that have been abandoned by an agent?
Response 19: Yes, error is the task that the agent gave up.
Point 20:- First sentence of the abstract.
Response 20: I have made changes in the text and made an annotation.
Point 21:- From introduction: "the ability of the agent to change during the execution of the mission may change" what does the first "change" refer to? change what?
Response 21: For example, the agent is destroyed. I have already modified this sentence in the paper.
Point 22:- pag 4: "the magnitude of \alpha is proportional to it", what does it stand for? "the difficulty and the priority" is plural.
Response 22: The description is inaccurate and is described in more detail in the latter sentence, so this sentence was deleted in the paper.
Point 23:- end of pag 4: "while the improved ... protocol in this study reduces mxn communication times"
- pag 5: "Rew represents used", "Cap indicated as".
- check "s" for plural and third person
Response 23: I have modified in the manuscript.
Round 3
Reviewer 3 Report
Authors addressed my concerns and modified the paper in order to meet them. The paper is more readable now, even if English form should be revised here and there and some sentences are still fuzzy. However, the protocol seems sound and results show it has benefits against competitors.
Some minor comments follow:
- pag 1: first sentence of abstract is still unreadable: what is the subject of the sentence?
- pag 4 line 135: please do not start a sentence with "And..."
- pag 4 line 142-142: C_true is the number of tasks successfully completed by the whole system up to that time?
- pag 4 line 143-144: as it is, the sentence states that "the agent is larger than Y", please restate.
- pag 4: do the lambdas, alpha and beta have default values? if yes please explicit them.
- pag 4 line 162: --> "reduces communication time to mxn". It is however weird not to know what is the starting point of such reduction (i.e., the communication time of the other protocols).
- pag 12 line 341-342: eliminate repetition and restate sentence
Author Response
Dear Editor and Reviewer:
Thank you for your letter and for the professor’s comments concerning our manuscript entitled “Task Assignment of Improving Contract Net Protocol under Multi-Agent System” (ID: 439235). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in red in the paper. The main corrections in the paper and the responds to your comments are as flowing:
Point 1: pag 1: first sentence of abstract is still unreadable: what is the subject of the sentence?
Response 1: This sentence has been modified.
Point 2: pag 4 line 135: please do not start a sentence with "And..."
Response 2: I have deleted "And".
Point 3: pag 4 line 142-142: C_true is the number of tasks successfully completed by the whole system up to that time?
Response 3: Yes, it is the amount of tasks when the final task is completely completed. It has been modified in the manuscript.
Point 4:- pag 4 line 143-144: as it is, the sentence states that "the agent is larger than Y", please restate.
Response 4: This sentence has been modified.
Point 5: pag 4: do the lambdas, alpha and beta have default values? if yes please explicit them.
Response 5: The initial values of lambdas 1 and lambdas 2 are 0.5, and the initial values of alpha and beta are both 0, which are already marked in the text.
Point 6: pag 4 line 162: --> "reduces communication time to mxn". It is however weird not to know what is the starting point of such reduction (i.e., the communication time of the other protocols).
Response 6: The number of communications that have joined the classic contract net protocol, and the number of communications that improve the contract net protocol.
Point 7: pag 12 line 341-342: eliminate repetition and restate sentence
Response 7: Duplicate content has been removed.
