=Paper=
{{Paper
|id=Vol-2173/paper12
|storemode=property
|title=Collective Story Writing through Linking Images
|pdfUrl=https://ceur-ws.org/Vol-2173/paper12.pdf
|volume=Vol-2173
|authors=Auroshikha Mandal,Mehul Agarwal,Malay Bhattacharyya
|dblpUrl=https://dblp.org/rec/conf/hcomp/MandalAB18
}}
==Collective Story Writing through Linking Images==
https://ceur-ws.org/Vol-2173/paper12.pdf
Collective Story Writing through Linking Images
Auroshikha Mandal, Mehul Agarwal and Malay Bhattacharyya
Department of Information Technology
Indian Institute of Engineering Science and Technology, Shibpur
Howrah – 711103, India
E-mail: malaybhattacharyya@it.iiests.ac.in
Abstract Related Work
Collaborative creativity is the approach of employing crowd Creative Crowdsourcing is currently a highly exercised con-
to accomplish creative tasks. In this paper, we present a cept, with many small start-ups using it to accomplish tasks
collaborative crowdsourcing platform for writing stories by and attract users. Platforms like DesignHill (DesignHill
means of connecting a series of ‘images’. These connected 2014) exploit the inputs from crowd workers to help design
images are termed as Image Chains, reflecting successive sce- logos for postings made by people. Another popular plat-
narios. Users can either start or extend an Image Chain by up- form SquadHelp (SquadHelp 2011) employs crowdsourcing
loading their own image or choosing from the available ones. to name products and ideas. Graphic designing is also done
These users are allowed to pen their stories from the Image using crowd inputs by the platforms like 99designs.com
Chains. Finally, stories get published based on the number
(99Designs 2008). However, these platforms work by select-
of votes obtained. This provides an organized framework of
story writing unlike most of the state-of-the-art collaborative ing only one from multiple inputs provided by the crowd
editing platforms. Our experiments on 25 contributors high- contributors. They essentially pick the best out of a pool,
light their interest in growing shorter Image Chains but voting with the crowd helping to fill that pool.
longer Image Chains. CorpWiki is a self-regulating wiki system for effective ac-
quisition of high-quality knowledge content from the corpo-
rate employees (Lykourentzou et al. 2010). However, such
Introduction platforms are not for creative tasks. Collabowriters is a plat-
Crowdsourcing involves using the power of crowd to per- form that turns crowdsourced inputs into novels. People are
form a task (Brabham 2008). The sheer power in involving allowed to enter lines consisting of a maximum 140 char-
the mass to distribute a job of big proportions makes this acters, and they are subsequently voted to decide the most
idea successful in performing various kinds of tasks, skilled popular one. The highest voted line is then added as the
or not-so skilled, technical or creative (Kittur et al. 2013). next sentence to a novel (Collabowriters 2012). This short
The aim of this study is to exercise the power of crowdsourc- lived project has tried to build a well written, coherent story
ing for carrying out tasks like collaborative story writing, of the size of a book, with the help of crowd. However,
and creative plot building, a field which can not be auto- we aim at making short stories at first, with the idea be-
mated by machines. As the people fill in their text descrip- ing to link creative thoughts together. There are also some
tions to make stories, we intend to record the input in the Wiki-based interfaces for collaborative story writing. One
form of creative links between story elements in the form of such platform asks students to edit on a common platform,
images (depicting scenarios). Like any crowdsourcing plat- with an interface like Microsoft Word, and builds new sto-
forms, this too thrives on the abundance of data. As the num- ries through posting a discussion (Hamid 2012). The users of
ber of people interacting with the interface increases, the ac- this platform have reported that the interface is not receptive
curacy, diversity and content on the platform also rises. To to multiple people editing a document simultaneously. This
employ this idea for creative plot building, we have primar- platform also suffers from the problem of content deletion
ily studied the existing collaborative editors and gained in- by the other users whenever a new story is being formed.
sights. This is finally used to design a platform that provides The users have also noticed the lack of an interactive way
an image based interaction. The stories are basically written to add ideas to a story. Some platforms (Storybird 2008),
through connecting images, termed as Image Chains. This (Inklewriter 2011), (Wattpad 2006) allow users to write a
creates a universal platform to merge together ideas of dif- complete story online on a platform, such that people can
ferent crowd workers. It has the capability to create growing view their stories. An online audience provides continuous
and evolving stories with time involving increased number feedback to the writers, helping them guide the story, and
of users and is, thus, a step toward organized story writing. also to improve the content. This in turn also provides read-
ers with a place to read new stories written by crowd work-
Copyright c 2018for this paper by its authors. Copying permitted ers. However, these sites work on adding complete stories,
for private and academic purposes. and are focused more toward an online platform to judge and
read new stories. We merge the working principles of plat-
forms like (Storybird 2008) and (Hamid 2012) to provide
users a place to get linked as well as vote for new stories. It
does not rely on people being expert story tellers, because
people have multiple roles to fulfill. So, most of these said
approaches are unorganized.
There are several crowd-powered models that serve the
purpose of organized creative writing too. Motif is a recent
platform which guides users through adding video snippets
from a journey or incident, and adding story-like descrip-
tions to each ‘scene’ they add (Kim et al. 2015). These are Figure 1: A snapshot of the page where people write their
joined together to form coherent stories. Motif thus gen- stories or vote for other stories.
erates good quality stories by inputs from novices and ex-
perts alike, by providing an organized platform for creation.
Another platform by Kim et al., Storia (Kim and Monroy- • Preservation of content: People can even delete each
Hernandez 2016), works to link social media updates about others’ inputs, and a lot of good ideas get wasted as a
an event to make a coherent story about a particular inci- result (log files can be ignored by others in the long run).
dent. The motive remains linking social media updates, but • Absence of role distribution: If there is no distribution
the approach involves asking the crowd to generate sum- of roles among the people, it leads to people overriding
maries from inputs. Storia hence takes the short social me- each others’ functions at any instant.
dia updates from Twitter, Facebook, etc. as nodes in the
story, which are to be linked to form a well written story. • Recency bias: Recent edits get more priority than the
A crowd-powered model by Kim et al., Mechanical Novel older edits.
(Kim et al. 2018), attempts at microtasking the 2 facets of • Arguments related to ownership and content deletion:
story-writing, choosing the target for a story and writing in- Since people can delete others’ inputs, it leads to unnec-
dependent scenes of the story, through mTurk. This paper is essary arguments between the collaborators.
focused on using the crowd to break down a high-level goal
such as creating a story into microtasks which can be self- Platform Design
managed by the crowd to fulfill/extend the primary goal. It A text-only platform initially seems like a good idea to con-
allows the crowd to decide on the current state of a story, and nect plots with the help of a crowd. However, as the size of
how it can be improved or added to. Then the crowd work- a story increases (the number of scenarios added to a story
ers propose the changes which should be made to a story, increases), the complexity of reading through already exist-
and these changes are voted upon by the others. We how- ing story elements to decide which ones to connect becomes
ever, allow people to merge two story paths together, and to higher. The lack of images makes it difficult for people to
branch one story into a completely new one. The addition of easily visualize what other people are creating, without hav-
text/task of writing text for an Image Chain, which finally ing to read through the whole paragraph. This gives rise to
becomes a piece of story, allows people to create what they the idea of an even more organized approach, and the abil-
feel is the best narrative for a given set of images. These ity to interact better with users. We have already pointed out
story pieces are voted by others to choose the best story for several limitations of existing approaches that led to the bet-
a given sequence. Mechanical Novel does not allow users to ter design of our platform.
continue a current story in a direction they want to, unless In our designed platform, a sequence of images which de-
the whole crowd decides on it. Our platform aims to pro- picts a flow of narrative is defined as an Image Chain. A
vide the flexibility of growing stories in any manner as users starting image from which such Image Chains are formed
want. is referred to as a Base Image. A crowd worker can either
start such a story (with a Base Image), continue a story (by
Motivational Insights extending an Image Chain), or write or edit a story (on an
The current paper basically aims at building a platform existing Image Chain), and finally vote for such a story (see
which allows users to add content to a creative story. There Fig. 1). All these steps, as listed hereunder, are however op-
are several reasons why a common document editing plat- tional.
form (e.g., Google Doc) will not serve the said purpose. The 1. Starting a story: A crowd worker can start a story by
human co-ordination can be managed by many existing plat- uploading an image with a description of it or by choosing
forms but the challenges remain to be the lack of an orga- an image already existing in the database.
nized structure, possible inclusion of noise, chaotic editing,
inconsistent results, etc. The main challenges that we have 2. Continuing a story: A crowd worker can continue a story
observed are listed below. by selecting a particular Image Chain (ordered chain of
images depicting a flow of events created by a crowd
• Lack of organization: If the platform is just an open doc- worker). Note that, the selected Image Chain also starts
ument which everyone could edit, there is a lot of chaotic with a Base Image. The crowd worker can either upload
input. an (or multiple) image(s) to continue or select an image
from the existing database of images. An uploaded image
by a user is added on to our pool of images immediately Table 1: Analysis of the length of Image Chains and votes
so that the next crowd worker can use the same as and obtained by them.
Average length of Image Chains 4.67
when required.
Average number of Image chains of length ≤ 5 5.5
3. Publishing a story: A crowd worker is entitled to write Average number of Image Chains of length > 5 2
a story based on the Image Chain he has formed. Ev- Average number of votes for a story text 3.18
ery crowd worker who has contributed to the same Im- Average votes for story texts for Image Chains of length ≤ 5 2.4
age Chains can write their stories by their own or take Average votes for story texts for Image Chains of length > 5 3.833
help from other contributors. Suppose a particular crowd
worker has written something about an Image Chain. Sub-
sequent crowd workers writing for the same Image Chain reason could be that the majority of crowd workers have ex-
can view what the former has written and use the insights tended 1-3 sized Image Chains and added 2-3 images more.
to create another version of the story. Now, the former Hence, even when a crowd worker is adding images to an
crowd worker can again use the insights of the latter to Image Chain of size > 7, the inclination is to extend it to
create a revised version of the story. 1-2 images more. Then these crowd workers would end the
4. Voting for a story: A crowd worker can select a particu- chain and start writing a story for the same.
lar Image Chain to vote from the set of all the story chains To ensure whether larger Image Chains obtain more votes,
formed till then. He can select from all the stories written we again compare the two groups of Image Chains (as listed
for that Image Chain and vote for his favorite. In this way, above, with threshold for chain length = 5). The mean and
a story is voted upon. Internally, votes for an Image Chain standard deviation values of votes obtained from the users
are assessed when a crowd worker creates an Image Chain for Image Chains are reported in Table 1. The comparison
already created by another crowd worker. In that case, in- of the two groups of Image Chains (segregated on the basis
stead of creating a redundant Image Chain, we increase of their lengths) was put to a t-test, which gives a significant
the number of votes for that Image Chain. Image Chains observation that longer Image Chains obtain significantly
with higher votes have a greater probability to be included higher number of votes (p-value = 0.0366; t-test). Hence
in the recommendation list. people are more inclined to alter or grow Image Chains of
shorter length (≤ 5, in our data), which gives us increased
Empirical Analysis concentration at the lower lengths, while people are opting
to vote more for images of longer lengths, may be because
Total 25 crowd workers (male = 16, female = 9, mean age they appear to be more complete as a story.
= 21.8 years) have taken part in the deployment session by
getting connected with computers and mobile phones. None
of them are by profession story writers or storytellers. Most Conclusion
of these people have used crowdsourcing platforms earlier, Content filtering is one of the primary concerns of a crowd-
albeit not knowing it is crowdsourced. They have used the sourcing platform. A system to filter the content, as well as
platform for 10-72 min (mean time of use = 45 min) in total. activity, on the platform needs to be present to ensure the
During this time, they have used the platform to add images, quality. The proposed platform attempts to do that by major-
build stories, and also give feedback about the use, interface ity approval and storing many versions of one Image Chain
and interest via a feedback form. From a starting pool of 30 with the argument that any chain can be extended later.
images (provided as Base Images), the platform has finally Additionally, the recommendation facility should be tuned
grown to 64 images at the end of experimental period of to the genre interest of the user. For this, image descrip-
about a month. Total 34 Image Chains (images selected by tions have to be categorized into buckets of similar tastes
the crowd workers depicting an ordered flow of thoughts) so that a user selecting images from one bucket is shown
have been formed and the users have contributed to 22 inde- images and Image Chains pertaining to the same or similar
pendent Story Texts. buckets (buckets with similar kind of genres). Recommen-
We have analyzed the Image Chains to study their aver- dation can also be provided based on the nature of contri-
age length (number of images they contain), and how likely butions. A crowd worker may extend the work of another
people are to extend chains of a particular length. The aver- crowd worker. Till now, the recommendation facility gives
age length of an Image Chain is found to be 4.67 after the importance only to the voting procedure. Highly voted Im-
experimental session, the maximum length being 11. To get age Chains and their corresponding texts are shown in the
an idea about whether users prefer to add images to (extend) recommendation section. A balance of votes, genres, con-
smaller chains or bigger ones, we have divided these Image tributors and the submission time of any story should make
Chains into two groups based on a length threshold value of a much better recommendation system. Better incentives are
5 (Since our average length was calculated as 4.67). Out of also an important concern here. We did not use any means
these groups, the average number of chains for lengths be- to incentivize the crowd workers except from providing an
low (<=) 5 images is found to be 5.5 and for lengths greater encouragement through a Leaderboard. Any such platform
than 5 images is found to be 2. Putting these two popula- would need some form of fund generation or fund collec-
tions under a t-test, we found them to be significantly dif- tion mechanism to financially support the competent crowd
ferent from each other (p-value = 0.0086; t-test). A possible workers.
Acknowledgement
This publication is an outcome of the R&D work undertaken
in the project under the Visvesvaraya PhD Scheme of Min-
istry of Electronics & Information Technology, Government
of India, being implemented by Digital India Corporation
(formerly Media Lab Asia).
References
99Designs. 2008. Logos, web, graphic design & more.
https://www.99designs.com.
Brabham, D. C. 2008. Crowdsourcing as a model for
problem solving: An introduction and cases. Convergence
14(1):75–90.
Collabowriters. 2012. Introducing the collabowrit-
ers – ideascale. https://ideascale.com/
introducing-the-collabowriters.
DesignHill. 2014. Graphic design website for logos, web
design & more. https://www.designhill.com.
Hamid, W. M. 2012. Discovering the potential of wiki
through collaborative story writing. The 8th International
Language for Specific Purposes (LSP) Seminar – Aligning
Theoretical Knowledge with Professional Practice.
Inklewriter. 2011. Write your own interactive
stories. https://www.inklestudios.com/
inklewriter.
Kim, J., and Monroy-Hernandez, A. 2016. Storia: Summa-
rizing social media content based on narrative theory using
crowdsourcing. In Proceedings of the 19th ACM Conference
on Computer-Supported Cooperative Work & Social Com-
puting, 1018–1027. ACM.
Kim, J.; Dontcheva, M.; Li, W.; Bernstein, M. S.; and
Steinsapir, D. 2015. Motif: Supporting novice creativity
through expert patterns. In Proceedings of the 33rd Annual
ACM Conference on Human Factors in Computing Systems,
1211–1220. ACM.
Kim, J.; Sterman, S.; Cohen, A. A. B.; and Bernstein, M. S.
2018. Mechanical novel: Crowdsourcing complex work
through reflection and revision. In Design Thinking Re-
search. Springer. 79–104.
Kittur, A.; Nickerson, J. V.; Bernstein, M.; Gerber, E.; Shaw,
A.; Zimmerman, J.; Lease, M.; and Horton, J. 2013. The
future of crowd work. In Proceedings of the 2013 confer-
ence on Computer supported cooperative work, 1301–1318.
ACM.
Lykourentzou, I.; Papadaki, K.; Vergados, D. J.; Polemi, D.;
and Loumos, V. 2010. Corpwiki: A self-regulating wiki
to promote corporate collective intelligence through expert
peer matching. Information Sciences 180(1):18–38.
SquadHelp. 2011. Company, brand & business name ideas
by experts. https://www.squadhelp.com.
Storybird. 2008. Storybird – artful storytelling. https:
//storybird.com.
Wattpad. 2006. Write on wattpad. https://www.
wattpad.com/writers/how.html.