Shailendra Rao - Stanford HCI Group

joinherbalistΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 6 μήνες)

60 εμφανίσεις

Page

1

of

4

Shailendra Rao

Abhay Sukumaran

CS 376 : Research in HCI

November 17, 2005

Project Milestone 2

Introduction

Online social networking is rapidly growing in popularity and usage. It has become
wildly popular in the United S
tates. The initial goal of our project was to build a social
networking website for the Business Process Outsourcing (BPO) community in India.
Because this community is closely knit and socially isolated from others because of their
odd work hours, we thou
ght that an online social networking platform would enhance
their social interactions.


Our preliminary research and discussions with people connected to the industry pointed
out a key flaw in our concept. Internet access for this user population is both l
imited to
particular sites at work and not as ubiquitous and always on as we are used to in the
United States. However, the overwhelming majority of BPO workers have cell phones.
This turned onto a project focusing on moving social networking to text messa
ging,
which is widely used in India.


Our project focus shifted once again after we reflected about multimodal interfaces in the
CS376 class readings and discussions. What if we could combine both the affordances of
text messaging with voice? Cell phones o
ffer both capabilities in a familiar and
ubiquitous manner. Our project has evolved into constructing a social networking tool
that runs entirely on the mobile platform, taking advantage of both its voice and text
messaging capabilities. This implementatio
n will offer a new paradigm for social
networking.


Hypothesis

We believe that our system can provide the affordances of a social networking site in the
form factor of a mobile phone. Additionally, incorporating voice into profiles will
provide a medium fo
r self
-
expression, which in some ways is richer than the current text
and picture profiles that online sites currently offer.


Evaluation Plan

We will test our claim that mobile phones offer a sufficiently rich interface to mimic
many of the key tasks of s
ocial networking websites with a usability study involving 7
-
15
Stanford students. Some of these key tasks include creating a profile, making changes to
it, looking up other members, etc. This lab
-
based usability study will allow us to observe
the actions
of our users while they use their phone (keypad and text messaging) to
navigate our voice user interface. We will record both the audio and video of our
participants interacting with our system.


Page

2

of

4

Creating a profile for the first time will be our most telli
ng task. The content of this
profile will be similar to those featured in popular web
-
based social networking sites such
as Facebook, Friendster etc. This task will test whether a voice input on a mobile phone is
sufficiently usable for creating a rich rep
resentation of oneself. Users will create a profile
complete with their demographic data and sound bytes describing their interests,
perspectives, hobbies, etc by going through a question
-
answer session.


Other tasks in our usability study will involve loo
king up a person by name, accessing
parts of their profile, and selectively editing their own profile. All these tasks will be
conducted on our prototype system, using voice, DTMF tones, and possibly text
messages. At the end of the study participants will

fill out a 15
-
item questionnaire that
covers attitudes towards the system, including its ease of use and their satisfaction with it.


The system is meant to be sufficiently robust and general that it can work with whatever
phones the participants might c
urrently be using, eliminating concerns of variance in
familiarity with any particular model of phone.


Current Prototype

The focus of our current prototype is on having a carefully
-
designed interactive voice
response system that allows the test user to cr
eate a profile. To access the system, the user
dials a standard number from their mobile phone. They then go through a series of voice
prompts. Since there is a social / dating context here, we use cross
-
paired voices


male
test subjects get female inter
viewers and vice versa. The conversation then proceeds, with
the user giving responses to the voice prompts and pressing # on their phones to indicate
the end of a response.


System: Hi (name), welcome to Goldspot social networking. Please use a short phr
ase to
describe what you want to do. You can say things like create profile, edit profile, find
person, add friend, change password.

User: create profile*

System: You are now going to create a profile. You will be asked several questions about
yourself. Yo
u can hang up at any time to cancel and your progress will be saved. Press #
to continue.

User: #

System: You will now be asked some demographic questions. Press # after you have
recorded each response. Are you male or female?

User: Female #

System: Ok, yo
u’re female. Pick an age range from the following …

.

.

System: Now let’s talk about your interests. What are your favorite movies?

User: My favorite movies tend to be romantic comedies, but I also love dark drama. For
example, I like Maid in Manhattan, bu
t I really love stuff like Eternal Sunshine of the
Spotless Mind. My other favorites are … #

System: OK. What about your taste in music.

Page

3

of

4

User: Classic rock’s the way to go! #

System: OK. What are some of your favorite books.

User: I’m not really that much
of a reader. I like to read some non
-
fiction though. Last
one I liked was this autobiography of Richard Branson.#

.

.

.

System: Thanks! You have answered everything required to create your profile. It will
now be available to your friends as determined by
your privacy settings. Have a great
day, and thanks for joining Goldspot!

---


* speech
-
to
-
text by Wizard of Oz techniques.

In this way the system assembles an audio profile of the person, in their own voice, which
can be accessed by other people. It is ak
in to an audio presentation of yourself, and the
interview format makes it easy for people to supply standardized categories of
information.


Further Development

There are several elements we plan to introduce into the prototype by the end of quarter.
They

might rely to some extent of Wizard of Oz techniques to do things like voice
-
recognition or speech
-
to
-
text. Our main focus will be on:


1.

Ability to look up a person by name, and get an audio representation of your
linkage to that person.

2.

Editing parts of y
our own profile selectively

3.

Accessing other people’s profiles

4.

Adding to a friend list


Some of these could involve short text messages. For example, since it is difficult for
voice
-
recognition systems to input names accurately, lookup could be done via a m
essage
sent a standard number, e.g.,


“Find Reena Vyas”


And the system would send you back a message which listed the people that matched and
their linkages to you.


Other possible goals could be


1.

Refine the user control of the voice interface. Allow th
em to do things like undo,
skip, re
-
record.

2.

Explore the possibilities of sending photos via MMS


Page

4

of

4

Related Work

Wired for Speech

by Clifford Nass and Scott Brave

Wired for Speech

will guide us in our selection of voice interface gender, tone, accent,
emotion
, and personality. It will also give us ideas on how to make our users feel
comfortable disclosing their personal information. Encouraging honest answers is another
thing we hope to get from this source. Our audio profiles are only interesting if a person’
s
real personality is showcased.


Voice User Interface Design by Michael Cohen, et al.

This book is billed as the definitive guide for VUI design. It will offer us both great
psychological principles to base our design decisions on as well as real world e
xamples to
learn from.


SCANMail:

a voicemail interface that makes speech browsable, readable and searchable

by Steve Whittaker, et. al

In essence the SCANMail team has converted Voice UIs to GUIs. Our system was not
intended to convert text in user’s prof
iles to voice (the opposite of SCANMail), but there
does seem to be some valuable evaluation insights that we can get from this paper.


Profiles as Conversation: Networked Identity Performance on Friendster

by Danah Boyd

One of the interesting points in th
is paper is how profiles on online social networking
websites such as Friendster have been used as communicative tools, rather than static
representations. One of our long term goals of this project is fostering narrative creation.
Because people always ca
rry their mobile phones, they can document interesting life
events at any time of their choosing. This will transform our system from a static profile
container to a more lively and active communication resource that is updated and
checked frequently. We a
lso aim to get hints of what specific portions of a profiles people
want to edit update frequently, so that we can provide easy access to this functionality.