Real-Time Filtering Non-Intentional Bid Request on Demand-Side Platform

Round 1

Reviewer 1 Report

Title: "Demand-side"--> "Demand-Side"

Abstract: demand-side platform (DSP)--> Demand-Side Platform (DSP)

List of Keywords: "DSP" may stand for "Digital Signal Processing". To avoid confusion, please replace it with Demand-Side Platform (DSP) in the list of keywords.

Introduction: Please consider an abbreviation for "Real-time Filtering Bid Requests" and use it in the rest of the paper.

Section 2: The title is "Online Advertising Ecosytems and Related Works". I think the authors had better separate these to topics "ecosystem" and "related works". In addition, I cannot see any ecosystem in this section. The ecosystem of of a technology must include enabling technologies, application areas, risks and threats, etc. I recommend that the authors use the existing literature to develop a real ecosystem for "Online Advertising". Moreover, there is room for a high-quality figure to demonstrate the established ecosystem.

Section 3: I suggest the title "Research Methodology" instead of "Dataset and Key Features".

Subsection 3.1: "the full real-world dataset which 191 was provided by TenMax AD Tech Lab Co., LTD" needs a reference.

Figure 10: The authors use to define the complete form of CDF before this figure.


Figures are of low quality, especially Figure 12.

The caption of Figure 11: "Real-time"-->"Real-Time".

The title of Subsection 6.2: "Results of Real-time Filtering Bid Requests process"--> Results of Real-Time Filtering Bid Requests Pocess".

Figure 14: Is there any possibility for replacing the figure with a figure with "English" writings (The same goes with Figure 15)?

Section 7: This section is not a good place to explain the detail of the method. I suggest that the authors focus on the achievements.

Most importantly, the authors should compare the evaluation results of their method with state-of-the-art methods and interpret the results obtained from the comparison.

Author Response

Please see the attachment.

Thank you!

Author Response File: Author Response.pdf

Reviewer 2 Report

The development of a method that can distinguish between intentional and non-intentional bid requests in RTB and determine, predict, or categorise them is a very essential and very relevant topic of research. However, the answer must to be demonstrated convincingly and ought to provide a strong foundational grasp of how the method of distinguishing deliberate and non-intentional bid requests in RTB is structured. My major concern on this paper is about measuring “Intent”. You should find another word to replace “Intension” based on many reason I highlighted below:

Due to the fact that the model building constitutes the bulk of the work, I have numerous concerns regarding it:

• What do you mean by "data pattern" This is something that needs to be thoroughly described because, in the context of data science, there are a great deal of terminologies that need to be operationally defined. This could generate confusion for the reader.

• You also lay forth the conditions for a threshold, but you do not define

• In the field of data science, I have never witnessed a situation in which a dataset spanning a period of 14 days (07/01/2018 192 – 07/14/2018) would produce any results that could be considered reasonable (this is my main criticism about the paper)

• How may "Intent" be measured using the bid request attribute (which includes information about the user and the publisher, such as time, audienceId, IP address, user agency, location, URL, host, domain,) as well as identification, measurement, and observation period?

• How might you identify "suspect audiences" that were responsible for generating fraudulent advertising traffic? You are utilising traffic analysis in this scenario; assuming that this is the case, how does traffic analysis determine "intent"?

• What are the approximate costs associated with the processing of the five processes outlined in the "3.1 Audience classification" section of your document?

• It is not quite obvious how data patterns and the threshold can be used to characterise bid requests as either purposeful or non-intentional. To begin, the variables "deliberate and non-intentional bid requests" are open to interpretation. Furthermore, within the context of "Intend," how can one calculate "Intend" using "data patterns and the threshold" without resorting to subjective evaluation? That does not bode well for evaluation of a genuine determination or choice regarding either the solution to the problem that falls under the purview of "Intend."

• Despite the fact that you brought attention to the fact that "1 flags = abnormal audiences and 0 flags as normal," the question remains: how do you construct the flag, and what are its constituent parts? How can "normal" and "abnormal" audiences be differentiated using the flag, and what are the theoretical justifications for doing so?

• The function that you claimed would build non-intentional bid request as well as intentional bid request is not clear, and you do not provide any details on how you come about making that function, nor do you provide some theoretical justifications. • You also do not provide some theoretical justifications. according to the structure of the

 

• How much of a cost burden would it be for you, in terms of computing costs, to process the five stages outlined in your section titled "5.1 Audience classification"?

Author Response

Please see the attachment.

Thank you!

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents a  Real-time Filtering Bid Requests model to predict whether an incoming bid request is intentional or non-intentional from the demand-side platform viewpoint. 

Intorudction section can be further improved by properly introducing the area i.e. what is real time bidding, what are the threats in real time biddings, different types of frauds, which type of fraud is under focus of this research work? 

my suggestion is to add/ merge  Online Advertising Ecosytems part of the paper to introduction section. 

The related work section should be a complete seperate section. Highlight and critically analyse existing detection systems, not just a simple overview. in related work section, many existing models are discussed, but not critically analyses, what are the limitations of existing models? why there is a need of new model (the proposed one?)

section 3.1 needs more details about datasets.

figure 2 may be improved for better readability. 

figure 11 is representing main concept, it needs more details of each module with respect to inputs and outputs.

Overall, the idea is fine, the results are convencing 

Author Response

Please see the attachment.

Thank you!

Author Response File: Author Response.pdf