MODERATION APPS FOR COMMUNITY REPORTERS IN RURAL AREAS Supervisor - - PowerPoint PPT Presentation

REALITY REPORTING AND MODERATION APPS FOR COMMUNITY REPORTERS IN RURAL AREAS

Mridu Atray 2008CS50217

Supervisor : Prof. Aaditeshwar Seth

1

Motivation and Problems

2

 Citizen Journalism – moderation indispensible

 Acts as a bottleneck when scaling such platforms

 Huge number of sources to access news in current

date

 Need to be able to link different news item together for

better comprehension

 Need to be able to visualize more than what meets the

eye

Objectives

3

 Design, build and validate a model for hierarchical

moderation as an extension to Goonj App

 Add features to Goonj App to better organize the

stories collected and enable better tracking of issues

 Design new web interfaces to make story

presentation more contextual.

JMV Media Flow

4

Community Reporter JMR Admin/ Moderator

5

Organization/Moderation Levels

Community Reporter JMR Admin/ Moderator Level 1 Level 2

System Flow for Moderation

6

Landscaping Study- Ranchi

7

 Field trip undertaken in August 2012  To understand mobile handling skills of the

community representatives

 Could complete menu driven browsing till a depth of 4-

5 levels

 All of them could make calls, send SMS but preferred

call over SMS

 Could not understand soft keys  Could not understand that there are more options which

are presently not on the screen

Central Moderation Data Analysis

8

 Nearly 1/4th items junk  Nearly 15% items need

guidance call, 7% audio editing

 Avg time spent in giving

guidance call - 32.33%

 Avg time spent in audio

editing - 31 %

Division of Responsibilities

9

Community Representatives

• Remove of junk

calls

• Give guidance

calls

• Give location,

topic, issue tags Central Moderation Team

• Write

transcripts for the audios

• Audio editing
• f the items

The Goonj Moderation App

10

Admin Interface to Delegate Items

11

Live Deployment - Ranchi

18-Jul-13

12

 In order to validate the designed model, moderation

app tested in the field

 Experimental setup, delegated items moderated by

JMV moderators as usual as well

Results - Accuracy

13

 Number of items needing guidance call = 33  Number of items publishable = 41  Overall accuracy = 45.95% < chance

Where is the confusion?

18-Jul-13

14

 Guidance call needing items

 (J + G)/G = 72.73% ( actual accuracy = 54.54%)

 Publishable Items

 (A+P)/P = 78.05% (actual accuracy = 39.03%) Categories Junk Call (J) Guidance Call Needed (G) Audio Editing Needed (A) Publishable (P) Guidance call Needed 6 18 2 7 Publishable 6 3 16 16

Inference and Possible Reasons

18-Jul-13

15

 Why are so many items needing guidance calls

marked as junk?

 Blank calls, someone attempting to record marked as

junk while should be guidance call

 Why are so many publishable items marked as

needing audio editing?

 Minor problems in the audio being recognized and

marked by creps

 Solution: Better training to identify differences

Community Rep Specific Analysis

18-Jul-13

16

 In order to ensure that lack of training is

actually the underlying problem

 Analysis conducted for each of the 3 creps

showed accuracy as

 crepA (current JMV moderator) = 69.23%  crepB (some past experience) = 37.5%  crepC (no past experience) = 29.17%

Decreasing previous moderation experience

Confusion Analysis for crepB and C

17

Results – Ease of Use of the App

18-Jul-13

18

 All delegated items moderated by each crep  Appreciated design considerations:

 To make a guidance call, just press “Call” button,

concerned number pre-assigned

 Ability to make guidance calls later, as low network

area and calls don’t go across several items

 While selecting location/topic/issue tags from list,

repeated selection not a problem, only last updated

Result – Distribution of Load

18-Jul-13

19

 Projecting theoretical benefit, (assuming 100 items

moderated per day)

 Average time to moderate single item = 7 mins  Time to moderate 100 items = 700 mins  Number of junk items =25  Number needing guidance call = 15  Items left to moderate = 100 – (25+15) = 60  Time to moderate these as usual = 420 mins  Time taken when tagging handled at level 1 = (0.8)*420 =

336 mins

 Time taken in new model as fraction of old = 336/700 =

48%

Challenges and suggestion

18-Jul-13

20

 Poor internet connectivity

 Built on flaky internet framework but can cause delay in

updates reaching the next level

 Long Power Cuts

 Lead to battery drainage  Tips to creps for better battery management

 Novice Users

 Issues such as “Message Memory Full” leads to phone

being blocked

Reality Reporting

18-Jul-13

21

 An issue once reported with JMV helps raise

awareness about the same

 As number of stories reported increases, it is

important that they don’t get lost in a big pile

 Need to identify related stories and

aggregate them as “Issues”

 Need to design and develop better web

interfaces to depict this better

Reality Reporting – Main Aims

18-Jul-13

22

 Under this subpart, we had the following aims:

 Design an admin interface for the moderators to be

able to create “issues”

 Allow these issues to be available on the moderation

app to allow linkage of stories to issues

 Design and develop interesting visual interfaces to help

communicate the insights from the collection effectively

Interface to create New Issues

23

 Added as an extra tab of the admin interface

to delegate items to creps

Addition to the Goonj App

24

 Design questions:

 Can creps also create a new Issue?  If yes, should issues created by one crep be synced to

• thers as well?

 How should the functionality be added to the app so as

not to overwhelm the creps?

Final Design

25

 Issue creation restricted to

central team

 Keeps check on the number

• f issues

 Removes the issue of sycning

between creps

 Very simple addition to app

as extra button to “Assign to an Issue” when adding location and topic tags

Web Interfaces - Timelines

26

Web Interfaces- Maps and Figures

27

Web Interfaces – Data Visualisation

28

Website

18-Jul-13

29

 For better visualization of the stories at JMV in

terms of Topic and Location

 Technical Specification

 Built on Django 1.4  Uses HTML5, CSS3  Deployed on Apache Web Server

 Currently accessible at : http://www.goonj.net

Screenshot – Main Page

18-Jul-13

30

Screenshot – Topic Channel

18-Jul-13

31

Screenshot – Reports & Statistics

18-Jul-13

32

Populating the Database- Automated Process

18-Jul-13

33

Goonj Website Server

Remote Server

goonjDB

URL

GET Request JSON file RSYNC Insert Values 1 4 3 2

Confusion Matrix for crepB and C

34

 crepB  crepC

Mridu Atray 2008CS50217

35

THANK YOU