Poqemon-QoE-Dataset

This work is proposed to describe a new dataset that contains many QoE Influence Factors (QoE IFs) and subjective Mean Opinion Scores(MOS). This dataset is collected during the PoQeMoN project (www.ip-label.fr/performance-wire/ projet-collaboratif-platform-quality-evaluation-mobile-networks) where a crowd measurement platform is implemented to assess YouTube end user’s QoE in mobile environments (UMTS, HSPA, LTE, etc.). This dataset concerns collecting a lot of QoE IFs using a VLC media player. The used platform consists on an Android-based applica- tion developed and installed on the end user device (phones and tablets). The testbed experiment mainly consists of the following elements:

• A dedicated mobile application has been developed for experimentation.

• The evaluation was performed at different locations (train station, walking, etc.).

• Users were trained to perform the video session tests.

• Nine devices are used by participants to achieve tests. The devices have different characteristics like screen size, android version and Cpu.

• Several types of videos were used (e.g. sport, movie trailer, documentary, news, music,... etc.).

• Four several mobile networks were tested (Orange, SFR, Bouyegues and Free)

1- Source:

Creators:

• Lamine Amour ([email protected])
• Sami Souihi ([email protected])
• Abdelhamid Mellouk ([email protected])

2- Data Set Information:

The dataset was built from a crowdsourcing campagn test where 1560 samples covering 29 Quality of Experience Impact Factors (QoE IFs). The used videos are of different types/complexities.

The geographical location of the testbed was the LiSSi laboratory (http://lab.lissi.fr/) around Paris city in France. 181 testers prticipated in the test campaign. All of them were researchers (their family) and students from different disciplines aged 19 to 38 years with few or no experience with video assessment experimentation.

3- Attribute Information:

-> The content of the dataset can be uploaded with different formats. These formats are:

name.csv : Microsoft Excel 2013 file.

name.arff : Weka files (https://en.wikipedia.org/wiki/Weka_(machine_learning).

name.data : Open Document format (https://en.wikipedia.org/wiki/OpenDocument.

-> Both files contain 22 QoE IFs and the Mean Opinion Scors (MOS)given by userss.

1. id

2. user_id

3. QoA_VLCresolution

4. QoA_VLCbitrate

5. QoA_VLCframerate

6. QoA_VLCdropped

7. QoA_VLCaudiorate

8. QoA_VLCaudioloss

9. QoA_BUFFERINGcount

10. QoA_BUFFERINGtime

11. QoS_type

12. QoS_operator

13. QoD_model

14. QoD_os-version

15. QoD_api-level

16. QoU_sex

17. QoU_age

18. QoU_study

19. QoF_begin_time

20. QoF_shift

21. QoF_audio

22. QoF_video

23. MOS

--> Each variable belongs to a category of QoE Influence Factors (QoE IFs) that consists:

• Video parameters from VLC video player (QoA) (3 to 10)

• Network information (QoS) (11 to 12)

• Device characteristics (QoD)(13 to 15)

• User's profil (QoU) (16 to 18)

• User feedback (QoF)(19 to 22)

--> Number of Instances :

-class 1 (MOS = 1): 92

-class 2 (MOS = 2): 119

-class 3 (MOS = 3): 244

-class 4 (MOS = 4): 787

-class 5 (MOS = 5): 300

--> Number of Attributes 23

--> NOTES: - 2st (Vresolution) attribute is class identifier for the video resolution (240p, 360p)

- 10st (Ntype) attribute is class identifier for the used network type in the assessment:

• 1 : EDGE

• 2 : UMTS

• 3 : HSPA

• 4 : HSPAP

• 5 : LTE

• 11st (Noperator) attribute is class identifier for the network operator in France:

• 1 : SFR

• 2 : BOUYEGUES

• 3 : ORANGE

• 4 : FREE

• 14st (QoU_sex) User's gender.

• 0 -> Woman

• 1 -> man

• 15st (QoU_study) High user's level study

• 5 -> University

• 4 -> Secondary school

• 3 -> College

• 2 -> Premary school

• 1 -> Other

• 19st (MOS) Mean Opinion Scroe that a tester will give at the end of each video view.

• 5 -> Excellent

• 4 -> Good

• 3 -> Fair

• 2 -> Poor

• 1 -> Bad

4- Related publications

1. Stéphanie Moteau, Fabrice Guillemin and Thierry Houdoin: Correlation between QoS and QoE for HTTP YouTube content in Orange cellular networks Conference 2017.

-> The data was used by Orange to identify the Quality of Service (QoS) indicators in order to predict the Quality of Experience (QoE) for HTTP YouTube content on mobile networks. A part of the results are based on experiments on the Orange network.

(2) Lamine Amour, Sami Souihi, Said Hoceini, Abdelhamid Mellouk: "Building a Large Dataset for Model-based QoE Prediction in the Mobile Environment". The 18th ACM International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems, pages 313-317. November 2015.

--> Here, the data set built methodlogy is explained, where the ACR method were used to evaluate subjectevly the user's QoE using Mean Opinion Score (MOS). Furthermore, this data was used to illustrate the QoE IFs interaction problem in the QoE assessment.

5- ACKNOWLEDGMENTS

This work has been funded by Orange in the framework of the French cooperative project “Platform for Quality Evaluation of Mobile Networks”, Pˆole de Comp´etitivit´e Systematic (FUI 16).