A Personalized Recommendation Algorithm Based on the User’s Implicit Feedback in E-Commerce

Round 1

Reviewer 1 Report

The paper entitled: " A Personalized Recommendation Algorithm Based on the User’s Implicit Feedback in E-commerce”, in general, the motivation of the paper is quite clear and rational. This paper tackles a very important problem and fitting well with the journal’s scope.

The paper has included interesting ideas and concepts; it is well grounded in the body of knowledge in the area. The paper is well written and organized and includes many mathematical definitions to support the approach followed. The approach followed looks useful and the results are promising.

However, In my opinion, the paper in the current version has to be refined at some important points:

•         What are the main limitations of this approach?

•         What are the managerial implications from this research.

•         How decision or policy makers could benefit from this study

·       It could be interesting to discuss in the conclusion part, the perspectives and the improvements of this work.

 

I would however encourage the author to pursue his efforts in improving the paper for future publication since the topic of the research is highly relevant.

Author Response

Dear reviewer:

Thanks for your advise, and the following are my responses.

Point 1: What are the main limitations of this approach?

Response 1 : This approach is very accurate on the time node requirements of the data. Because, we need to  split the user's behavior log into different purchase cycles. And the approach is designed for E-Commerce  and need to be changed if used in other areas.

Point 2: What are the managerial implications from this research?

Response 2:  The goal of this research is to predict what the user will buy next, thereby reducing the time users spend looking for items.  If the user can find the desired item as soon as possible, it will greatly increase the transaction rate. And I write the benefit of a good recommendation algorithm for the e-commerce platform at introduction.

Point 3 : How decision or policy makers could benefit from this study?

Response 3: Decision or policy makes apply good recommendation algorithms to their e-commerce platforms, enabling their e-commerce platforms to better understand users and increase user viscosity.

Point 4 :  It could be interesting to discuss in the conclusion part, the perspectives and the improvements of this work.

Response 4: I add some discuss in the conclusion part. 

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors present an very relevant approach of a personalized recommendation system based on an algorithm that use the feedback and log of users. The approach is well introduced, the algorithm is very well formalized and mathematically explained. Perhaps the state of the art should be improved with more references and using other more basic but relevant papers for instance:

 - J. Bobadilla, F. Ortega, A. Hernando, A. Gutiérrez, Recommender systems survey,

Knowledge-Based Systems, Volume 46, 2013, Pages 109-132, ISSN 0950-7051,

- Burke, R., Hybrid Recommender Systems: Survey and Experiments, User Model User-Adap Inter (2002) 12: 331. https://doi.org/10.1023/A:1021240730564

The experiments and discussion of the approach comparing with other methods are very significative and the results demonstrate a proven advantage of the proposal among others.

Minor issues:

- Include an space between word-reference or word-brackets.

Author Response

Dear Reviewer:

Thanks for your advise, and the following are my responses.

Point 1:  Perhaps the state of the art should be improved with more references and using other more basic but relevant papers for instance

Response 1:I have read the two papers, they all summed up the recommendation algorithms for different periods. With the evolution of RS，the RS tend to make use of different sources of information (collaborative, social, demographic, content, knowledge-based, geographic, sensors, tags, implicit and explicit data acquisition, etc.), and I have discussed the improvements of our approach in the conclusions. In the future, we will combine the user behavior log with the user's demographic features (age, gender, and occupation) and the item’s features information (brand, category) .

Author Response File: Author Response.pdf

Reviewer 3 Report

(1) In Figure 4,5,6, the precision values and recall values among compared methods are too low like 0.02~0.08 and 0.05~0.20. From your results, it is hard to say that you proposed method is efficient method. Why the values are too low? Is it meaningful?

(2) Show some recommended examples in e-commerce using your proposed method. User satisfication is an important issue in recommendation system.  Low precision means low satisfy. Also, recall means low filtering from user's request. Please show the user's satisfy degree in any ways.  

Author Response

Dear Reviewer:

Thanks for your advise, and the following are my responses.

Point 1 : In Figure 4,5,6, the precision values and recall values among compared methods are too low like 0.02~0.08 and 0.05~0.20. From your results, it is hard to say that you proposed method is efficient method. Why the values are too low? Is it meaningful?

Response 1: There are two reasons for the values are too low. First, our method analyzes the user behavior log, which  lacks rating information. We can't directly obtain user's preference, which has a certain impact on the calculation of user similarity. Second, the test data only include 5 days of JData, and the number of items purchased by a user is little.

Point 2: Show some recommended examples in e-commerce using your proposed method. User satisfaction is an important issue in recommendation system.  Low precision means low satisfy. Also, recall means low filtering from user's request. Please show the user's satisfy degree in any ways.  

Response 2 : I am sorry that our approach is in the experimental phase and has not been used for a real e-commerce platform.So there is no way to judge user's satisfaction.But the data we experimented with were all from the real data of JD.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

1. Please add user's satisfication degree or real using data from samle users in anyways. That is optional accetance.

Author Response

Point 1 : 1. Please add user's satisfication degree or real using data from samle users in anyways. That is optional accetance.

Response 1 : We gave an example with a single user and drew a table showing the recommended results for the different methods in the end of my paper.

Author Response File: Author Response.pdf