Feedback

HEC-Ecole de gestion de l'Université de Liège
HEC-Ecole de gestion de l'Université de Liège
MASTER THESIS
VIEW 142 | DOWNLOAD 578

Investigating the effect of volume, rating, and valence of product reviews on helpfulness

Download
Carnevali, Gauthier ULiège
Promotor(s) : Ittoo, Ashwin ULiège
Date of defense : 6-Sep-2016/12-Sep-2016 • Permalink : http://hdl.handle.net/2268.2/1892
Details
Title : Investigating the effect of volume, rating, and valence of product reviews on helpfulness
Author : Carnevali, Gauthier ULiège
Date of defense  : 6-Sep-2016/12-Sep-2016
Advisor(s) : Ittoo, Ashwin ULiège
Committee's member(s) : Beretta, Alessandro ULiège
Ghilissen, Michael ULiège
Language : English
Number of pages : 72
Discipline(s) : Business & economic sciences > General management & organizational theory
Institution(s) : Université de Liège, Liège, Belgique
Degree: Master en sciences de gestion
Faculty: Master thesis of the HEC-Ecole de gestion de l'Université de Liège

Abstract

[en] The purpose of this thesis is to try and create a model capable of predicting if a review will be helpful or unhelpful.
The data that was used comes from Amazon. Four samples were used and were divided into two major groups: experience goods and search goods. The two experience goods samples were video games and books while the two experience goods samples were electronics and cell phones.
I identified three main independent variables which were: polarity of the text, difficulty of the text, and number of star ratings given by the reviewer.

I then hypothesized how these variables would react according to whether we were dealing with experience or search goods.

Experience goods:

• : Reviews with less star ratings are more helpful.
• : Texts with a negative or positive polarity are more helpful.
• : Reviews with more difficult texts are more helpful than easy reviews. However, reviews with extreme text difficulties are less helpful.

Search goods:

• : Reviews with higher star ratings are more helpful.
• : Neutral texts are more helpful than reviews with a positive or a negative polarity.
• : Reviews with a text difficulty of plain English, fairly easy, and fairly difficult are more helpful.

The next step was to create a model capable of predicting helpfulness, for this I used a binary logistic model. I tested for assumptions. All assumptions were met apart for assumption #7: “there shouldn’t be any significant outliers”. All samples had outliers, however I decided to keep them in the final models, they could be outliers because of a lack of independent variables. They had studentized residuals ranging from 2 to 3.
All the models with explanatory variables had an increase between 10 and 20%.

For the search goods models and the video games sample model the variables contribution when in this order: rating, difficulty, and polarity. With rating being the most important. For the books sample the difficulty of text was more important than the rating. The video game sample seemed to follow the same pattern as search goods.
The polarity variable contributed to the model, but not as significantly as the others, changing this variable or improving the sentient analysis could prove to increase model precision.

I conclude by saying that, categorizing products into two large categories isn’t enough to identify what exact variables influence products. I also write about possible practical applications of this analysis. Such as:
• Improved ranking system of reviews
• Help reviewers know what are the criteria for a helpful review, and so improve review helpfulness in general
• Help manufactures improve their product and gain insight on customer demand.


File(s)

Author

  • Carnevali, Gauthier ULiège Université de Liège > Master sc. gest.

Promotor(s)

Committee's member(s)

  • Total number of views 142
  • Total number of downloads 578










All documents available on MatheO are protected by copyright and subject to the usual rules for fair use.
The University of Liège does not guarantee the scientific quality of these students' works or the accuracy of all the information they contain.