GeoPoetry: Designing Location-Based Combinatorial ... - Springer Link

12 downloads 0 Views 261KB Size Report
Spotify, using a simple model of affect. The result is a sonic reflection of the social geography traversed by the user that responds to its situatedness in both ...
GeoPoetry: Designing Location-Based Combinatorial Electronic Literature Soundtracks for Roadtrips Jordan Rickman and Joshua Tanenbaum ✉ (

)

Department of Informatics, University of California, Irvine, Irvine, CA, USA {jrickman,joshua.tanenbaum}@uci.edu

Abstract. In this paper we present GeoPoetry, a location-based work of elec‐ tronic literature that generates poetic language and dynamic soundtracks for roadtrips that reflect the mood of people in the surrounding area. GeoPoetry takes recent nearby geotagged Twitter data and generates strings of combinatorial poetry from them using simple Markov-chain text generation. It also performs a sentiment analysis on the local Twitter traffic, which it uses to seed a playlist on Spotify, using a simple model of affect. The result is a sonic reflection of the social geography traversed by the user that responds to its situatedness in both space and time. GeoPoetry participates in a long tradition of public and locative artwork which has the potential to inspire exciting new works of interactive narrative. Keywords: Locative media · Electronic literature · Combinatorial narrative · Sonification · Emergent narrative · Markov chains · Sentiment analysis

1

Introduction

As mobile computing grows more pervasive and ubiquitous a small (but growing) collection of artists, researchers, and designers have started to create playful digital experiences that are meant to be experienced away from the stationary desktop or gaming console. The conceptual roots of these traditions lie in a number of interconnected 20th century artistic movements that sought to liberate art from authoritarian systems and contexts including Dadaism [1], Fluxus [2, 3], Situationism [4, 5], and Happenings [3, 6], as well as avant garde and political theater practices such as Boal’s Theater of the Oppressed [7] and Brecht’s Epic Theater [8]. These movements subverted existing orthodoxies around art, power, and the notion of the audience or spectator, often proposing and executing performances and artworks that undermined any critical or conceptual distance between artist, viewer, and context. They form a conceptual foun‐ dation for contemporary location-based digital entertainment systems, which can trace their immediate technological and theoretical roots to a number of parallel academic traditions including Ubiquitous Computing [9, 10], Pervasive Gaming [11], and Context-Aware Computing [12–14]. Connecting all of these areas is a broad concern with location awareness, mobi‐ lity, and sensing technology to attends to the situatedness of the user in time and space. Within the Interactive Digital Narrative (IDN) community, there has been an increased interest in these kinds of pervasive and locative technologies as a platform © Springer International Publishing AG 2016 F. Nack and A.S. Gordon (Eds.): ICIDS 2016, LNCS 10045, pp. 85–96, 2016. DOI: 10.1007/978-3-319-48279-8_8

86

J. Rickman and J. Tanenbaum

for adaptive storytelling technology, but the design space for these experiences remains underdeveloped when compared to the design space of more traditional, nonpervasive, digital narratives. In this paper we present GeoPoetry: a new combinato‐ rial location-based work of electronic literature that expands the design poetics of pervasive storytelling into new territory. We describe our vision for GeoPoetry – currently a work-in-progress – and then describe the design and implementation of the current system. We reflect upon our design process which has illuminated a set of challenges and design considerations that we consider instructive of other more fundamental poetics for locative narrative technologies.

2

Key Concepts and Previous Work

In their book on Pervasive Games, Montola et al. argue that one of the defining char‐ acteristics of the medium is an expansion of the bounds of Huizinga’s “Magic Circle” [15, 16] along three dimensions: spatial, temporal, and social [11]. Pervasive games don’t take place in a fixed location or playing field, they are not bounded by traditional play-times or crisply delimited play sessions, and they take place in public spaces, often with-and-around spectators and other people who are not explicitly playing the game. For this work we attend specifically to the notion of “spatial expansion”, which Montola et al. further break down into three different categories: location-free games, site-adapt‐ able games, and site-specific games [11]. • Location-Free Games are games that use space and distance to expand gameplay, but which are indifferent to the specifics of the spaces where they take place. Thus, a location-free game might require a player to deviate from a normal route, or to seek a high-vantage point, but these activities could be transplanted to another setting without meaningfully altering the game. • Site-Adaptable Games incorporate some semantic awareness of their location into their gameplay, often at the level of generic categories that can be found in a variety of locations (such as banks, parks, schools, etc.) A site adaptable game can be shifted from one region to another so long as the new location has a similar complement of these generic settings. • Site-Specific Games draw on the specific historical, aesthetic, and architectural details and affordances of the location where they are designed and deployed. Because they incorporate specific local details into their design they cannot be relo‐ cated without significant cost. In our review of existing locative digital storytelling systems, we found that the line between “location-free” and “site-adaptable” designs was blurry, with most designs being either site-adaptable or site-specific. For this reason, we choose to focus on these two categories.

GeoPoetry: Designing Location-Based Combinatorial Electronic Literature

87

2.1 Site Specific Storytelling Systems The vast majority of current locative storytelling systems are designed to explore a specific site. A common strategy is to create an adaptive “guide” for a museum or other cultural heritage site. Lombardo & Damiano created a location aware animated character who supported visitors exploring the historical Palazzo Chiablese in Turin, Italy [17]. Wakkary et al. developed Kurio: an interactive museum guide that tracked the locations of family groups as they explored a local history museum in Surrey, Canada [18]. Lu and Arikawa created a map-based storytelling tool that arranged events along a walking tour path through Tokyo, Japan [19]. In Riot! the lines between interactive heritage guide and interactive drama are blurred through the use of site-specific mobile technology to recreate important moments from a series of 19th century riots in Britain [20, 21]. Other site specific systems are closer to distrib‐ uted theater, like Hansen et al.’s Mobile Urban Drama where groups of live actors are staged around a large environment that the spectator/user navigates using a mobile phone. Paay et al.’s Who Killed Hanne Holmgaard? is a similar, but more technologically mediated experience, where pairs of players navigate the city center of Aalborg, Denmark, uncovering pieces of the fictional world on their PDA devices [22]. This fits into a paradigm of locative storytelling where the environment becomes the site of a mediated scavenger hunt. There have been several other designs where media is distributed through space including Nisi et al.’s HopStory [23] and Location-Aware Multimedia Stories [24] and Crow et al.’s M-Views [25]. 2.2 Site-Adaptable Storytelling Systems There are fewer site-adaptable systems to be found, but they are more directly relevant to the design of GeoPoetry. One direct precedent for our work is Backseat Playground [26, 27] which is designed to be played by passengers in the backseat of a moving vehicle. Backseat Playground incorporates local GIS information about nearby land‐ marks and common geographical objects into a sequential narrative that is experienced over the course of a journey. Another inspiration for GeoPoetry is Schoning et al.’s WikiEar system, that draws on crowdsourced text to generate narratives about specific locations [28, 29]. Similarly, the GEMS system allows users to upload personal narra‐ tives about particular locations, relying more heavily on crowdsourced data [30]. GeoPoetry is site-adaptable, but it is unique in that it mines real-time data from its surroundings. In the following sections we describe it in greater detail.

3

System Design

The current prototype of GeoPoetry is still very much a work-in-progress, but we believe that in its current state it still serves as an interesting exploration of the design-space of location based narrative systems Our current implementation comprises two compo‐ nents: a web frontend and a server backend. The frontend is written in HTML, CSS, and Javascript using the popular AngularJS framework. The backend is written in Python. It

88

J. Rickman and J. Tanenbaum

depends on several open-source Python projects, namely: Flask, a web microframework1; Flask-CORS2, a Flask extension adding support for Cross-Origin Resource Sharing; Twython, a wrapper for the Twitter API3; Spotipy, a wrapper for the Spotify API4; pytest, a unit testing framework5; fudge, a framework for mocks and stubs6; markov-text, a Markov-chain text generator7; and VADER, a rule-based sentiment analysis library8. The backend and fronted communicate over HTTPS using the JSON data serialization format. Each request to generate location-linked poetry requires 5 parameters: latitude, longitude, radius (specified as either miles or kilometers), genre, and energy. First, the Twitter API is queried for tweets that are geotagged within the radius of the specified latitude and longitude. Some simple filtering is applied in an attempt to exclude marketing and promotional tweets – see the discussion section below. The tweet text is cleaned of URLs, hashtags, and “@mentions.” The resulting corpus of text is fed into a Markov-chain text generator, which generates a few lines of “poetry” that are, statisti‐ cally speaking, similar to what people in the area are saying on Twitter – albeit largely nonsensical. Next, this same corpus of text is fed into VADER’s sentiment analysis algorithm. Sentiment analysis yields a parameter called the valence, which ranges from −1 (extremely negative affect) to 1 (extremely positive affect), and can be any number in between. Spotify’s music recommendation API requires at least one seed, which can be a genre, track, or artist. We use a seed genre, which may be selected by the user but defaults to “ambient.” In addition to the genre, we specify three parameters for Spotify’s API: valence, given by our sentiment analysis; energy, a measure of activity and inten‐ sity; and instrumentalness, which we always set to its maximum value. This is to prevent lyrical content from the soundtrack to interfere with the poetic output of the text gener‐ ator. The Spotify API returns a track ID, which the web frontend uses to display a playable Spotify widget alongside the generated lines of poetry. We employ Russel’s “cicumplex” model of affect as a simple model for the concep‐ tual “mood space” for this work [31]. Russel divides affect into two axes: valence and arousal, where valence is positive vs. negative affect, and arousal is high vs. low energy. This model lends itself well to describing music, in particular, and maps well to param‐ eters within Spotify’s API. Valence and arousal together describe the mood that the music track seeks to capture and convey. However, as we discuss in more detail in the next section, VADER’s sentiment analysis only yields one dimension of affect: valence. To determine arousal, we draw on GPS information about the speed of the vehicle: faster speeds yield higher arousal, while slower speeds yield lower arousal. A key feature of the system remains in development: speech synthesis. In order to properly experience the poetry generated by the system while driving, the poetry needs 1 2 3 4 5 6 7 8

http://flask.pocoo.org/. https://github.com/corydolphin/flask-cors. https://github.com/ryanmcgrath/twython. https://github.com/plamere/spotipy. http://pytest.org/. http://farmdev.com/projects/fudge/. https://github.com/codebox/markov-text. https://github.com/cjhutto/vaderSentiment.

GeoPoetry: Designing Location-Based Combinatorial Electronic Literature

89

to be read aloud. This feature is currently being implemented, and there are several design considerations that have already become evident through the process. First, the design needs to consider the two different time scales at work within the system. Song lengths can vary significantly but are often between 3 and 8 min in duration. The reading aloud of a fragment of GeoPoetry takes significantly less time, with most of the poems generated by the system lasting less than 30 s. Second, there is a probability of textual confusion if the song queued by the system includes lyrical content which could render the text of the poetry incoherent. Finally, there is the aesthetic question of how to best juxtapose the reading of the poetry alongside the playing of the music, and the technical question of how to manage the volume level of both of these such that both are audible and intelligible. To solve the first problem, we propose that GeoPoetry only generate and read new poems when a new music track is triggered. We believe this will strike a balance between textual content, and musical content, and will also allow the system to re-sample the current location, yielding poetry that moves with the listener through both space and time. We have already addressed the second issue, by using Spotify’s API to only search for instrumental (non-lyrical) music. This serves our aesthetic goals for this work by creating playlists that reproduce some of the poetics of film scoring. The third challenge for this design remains unresolved, and will require significant trial and error in order to produce aesthetically pleasing and intelligible combinations of words and music. System Architecture. Figure 1 illustrates the architecture of Geo-Poetry, and the flow of information between components of the system. Dashed lines indicate information passed in an HTTP request, and dotted lines indicate information returned from an HTTP request. As described earlier, “cleaned” tweets have all hashtags, @mentions, and URLs removed.

Fig. 1. Architecture diagram illustrating data flow when generating poetry

90

J. Rickman and J. Tanenbaum

Sample Output. For the purposes of testing the geographical situatedness of the output, we ran the system with a variety of GPS coordinates. Table 1 shows some of the gener‐ ated “poetry”, and the valence used to choose a song. All requests included in the table used a radius of ten kilometers, and were performed on the evening of Friday, August 18th (Pacific Time). Note that all valences range from 0.0 (extremely negative) to 1.0 (extremely positive). During this testing process, we were struck by the preponderance of material that scored negatively on the affect scale. This confirms for us the common sense that twitter is an overwhelmingly negative place9 and raises some important questions for us for future work. Table 1. Sample output at multiple locations Location UC Irvine, California

Kuhio Beach, Honolulu The Capitol Building, Washington, D.C.

Taksim Square, Istanbul

The Vatican, Vatican City

Dome of the Rock, Jerusalem

King’s Cross Station, London

9

Poetry I love is Friday and I’m kinda tired but for a message - mr robot tho - at Library on the beginning again is gorgeous Yes stop - I instantly think they live in Mexico and Insta - Designed by marine cops Body hurt chest hurt I’m probably just my one name and us raise funds for helping import cholera epidemic to 658 of cartoons*Whacks*shoves at - They Luck When Sh_t Get Real on my profile - Gotta get more on television by and retweeting on this African Mechanic out Alma Mazlumun Ah n g n ergenekon - Iktidar g nayd n Anadolu Hisar in stanbul - Bu ehir yaln zca beni de il passe cr me about burkinis tho U can get Pure io oggi non meno di R diger e quando pensiamo che prenda decisioni stupide - Car By B - There are ten of 69F Menpar Arief Yahya Berencana Hadiri Toba Grand Fondo 2016 Thirsty Check out with a true Sigh - Least of the time - Whats love I guess ChooseLoveAlbum OutNow That’s why not on arrival crews assisted a quiz on showing this year then Olympics mixed with my - ah…you’re like blood donation - I just finished running 6.36 km in the palm trees

Valence 0.091233

0.100336 0.098348

0.021440

0.078147

0.041793

0.147892

This sentiment is common within the popular media: https://www.buzzfeed.com/charlie‐ warzel/a-honeypot-for-assholes-inside-twitters-10-year-failure-to-s?utm_term=.wmwA54Lx B#.jjBj5KP6o.

GeoPoetry: Designing Location-Based Combinatorial Electronic Literature

4

91

Discussion and Design Considerations

In designing and developing GeoPoetry we have encountered a number of challenges that we think are of relevance to others developing location-based storytelling systems, especially once real-time data becomes involved. In this section we consider these difficulties and discuss how we’ve addressed them in our design. 4.1 Neutral Valence A key element of our vision for GeoPoetry is that the system should capture the overall mood of the crowd in the area. In our current implementation, this feature is provided by the sentiment analysis component. The sentiment analysis library that we use, VADER, is specifically tailored for analyzing social media content [32]. However, gauging the overall sentiment of the crowd proves to be a more challenging problem, thanks to dynamics of scale. In order to properly seed the Markov-chain text generator, our system consumes a large number of tweets – the current implementation reads 500 tweets, although that number is a configuration constant and can be easily changed. Sentiment analysis is applied to each tweet individually, and the resulting valence measure is averaged across all tweets read. At such a large scale, the valence falls prey to the law of averages – average valence tends towards neutral (zero). Aesthetically, this is a problem: it yields uninteresting and uniform results when plugged into our system. In an attempt to miti‐ gate this problem, we experimented with excluding relatively neutral tweets from the average. We considered tweets to be “relatively neutral” if their valence fell within a certain interval centered around zero – for instance, plus or minus 0.2. However, with so many data points, positive and negative tweets tended to cancel each other out, and the problem persisted. We never saw the average sentiment deviate more than 0.35 away from neutral (on a scale from −1 to 1). These results may indeed reflect the overall sentiment of the crowd – it is intuitive that given a large number of people not necessarily discussing the same topic, their average sentiment will tend toward neutral. Nonetheless, we desire a system that will provide more variety in valence, and we want assurance that significant spikes in overall sentiment – for example, in the event of a breaking news story – will be noticeable by the listener. We will explore this challenge further in future iterations of the project. A couple possible approaches stand out. A simple solution might be to process fewer tweets. More variability in average valence can then be expected. However, experimentation would be needed to find an ideal corpus size that balances the need for fewer tweets in order to gauge sentiment with the need for a large corpus in order for the Markov generator to recombine the text into a sufficiently original composition (given too small of a corpus, such a generator will often regurgitate its input verbatim). Alter‐ natively, we may be able to apply more sophisticated statistical techniques to detect interesting trends or outliers.

92

J. Rickman and J. Tanenbaum

4.2 One-Dimensional Sentiment Analysis Most implementations of automatic sentiment analysis are one-dimensional. That is, they return a single measure of valence: from negative, through neutral, to positive [32]. While this is useful information, emotion is clearly a far more complex and nuanced thing. Moreover, in GeoPoetry we are specifically interested in mapping affective infor‐ mation to musical accompaniment, and the affective qualities of a piece of music are also multi-dimensional. We have already described how we are specifying both “valence” and “arousal” for a song selection, and how the one-dimensionality of sentiment analysis forces us to specify song arousal independently of the set of tweets the system reads. Spotify’s recommendation API accepts several more parameters that might relate to the perceived affect of a track, including “acousticness,” “danceability,” “loudness,” and “tempo.” Other recommendation engines may accept more or different parameters. One-dimen‐ sional analysis and mapping of “valence” only scratches the surface of the general chal‐ lenge of selecting music that matches the affective content of a textual corpus. Fortunately, there is some research in the direction of multi-dimensional sentiment analysis. SenticNet [33] is a project aimed at developing sentiment analysis based on semantic concepts, as opposed to individual words. The project provides a database of 30,000 (English-language) concepts, accessible via a REST API. Each concept is described by four dimensions: “pleasantness,” “attention,” “sensitivity,” and “aptitude,” as well as a “polarity” measure intended to be closer to the typical one-dimensional valence. The authors describe a new model of affect, the hourglass model, using the four dimensions [34]. Although we discovered SenticNet while developing our current implementation of GeoPoetry, we did not have the resources to integrate it. The API provides only a data‐ base of concepts, so we would need to implement an algorithm to recognize the concepts contained within a text, as well as an algorithm to combine those concepts into some representation of overall sentiment. We did not have the time to tackle such a nuanced affective computing problem, so we resorted to one-dimensional sentiment analysis, for which we found open-source implementations that could be readily integrated into our codebase. In future iterations of the work, we may explore the use of SenticNet, or other multi-dimensional sentiment analysis frameworks. Sentiment analysis is an active area of research in affective computing, so the future holds the promise of a more sophisti‐ cated mapping from source text to affectively relevant music. 4.3 Filtering for “Everyday Human” Tweets Our vision for GeoPoetry is that it should reflect the sentiment and conversation of the nearby crowd, by which we mean everyday human beings living or traveling in the geographic area. Twitter is an ideal source corpus for this, because of its sheer popularity, its widespread integration of geotagging capability, and its microblogging format (which encourages frequent updates, giving us a constant stream of new material). However, Twitter’s status as the dominant microblogging platform also creates a challenge to our vision: the proliferation of tweets that do not contain the remarks of an actual human

GeoPoetry: Designing Location-Based Combinatorial Electronic Literature

93

being. In particular, Twitter is a popular platform for marketing and branding. Therefore, we apply several filters to the set of tweets our system reads, in order to limit our corpus as best we can to the remarks of everyday people. First, we excluded “retweets.” Retweets are often a vehicle for sharing third-party content, not one’s own remarks. Even if the tweet being retweeted is the remark of an everyday person (say, the retweeter’s friend or acquaintance), including retweets will create duplication in the corpus. Second, we exclude tweets from “verified” accounts. Verified accounts are given that special status by the Twitter corporation, in a process that is opaque to outsiders. According to the Twitter FAQ, “Verification is currently used to establish authenticity of identities of key individuals and brands on Twitter.”10 Verified accounts are often celebrities or other public figures, or marketing accounts belonging to corporations. Therefore, their content does not reflect the thoughts of everyday human beings. Following a similar logic, we exclude tweets from accounts with more than 10,000 followers, reasoning that someone with such a large following is unlikely to be an “everyday human being” - or at least, is likely to tailor their content to their large following and personal brand, reducing the authenticity of their remarks. Although we have no quantitative measure of the success of these filters, they seem to be effective enough for our purposes. The most important measure of success was that our system did not produce poetry that sounded like it was recombined from an advertisement. With Markov-chain generation on a large corpus, a few not-regularhuman tweets can slip through without significantly affecting the results. Nonetheless, there may be room for improvement in future work. For example, one could argue that the duplication provided by retweets is desirable, provided the original tweet is the remarks of an everyday human. The resulting “signal boost” would make the retweeted content more likely to come out of the Markov text generator, reflecting the fact that multiple people saw it as a remark worth sharing. The selection of which tweets to exclude or include is a matter of authorial discretion, and we will continue to experiment with it.

5

Conclusions and Future Work

In this paper we’ve presented GeoPoetry – a new approach to location-based storytelling, drawing on real-time social media data to create combinatorial soundtracks to road trips. While the system itself is still clearly a work-in-progress, we believe that the design and development process have already uncovered some interesting and useful design poetics that have not be discussed within the electronic literature and digital storytelling communities. There is much work still to be done on this project at both the technical and artistic levels. Technically, we have identified a number of features that need to be implemented in order to consider the system complete, the most important of which is the inclusion of a speech synthesis capability. However, the more interesting questions surrounding 10

https://support.twitter.com/articles/119135.

94

J. Rickman and J. Tanenbaum

this work have to do with the experience that it creates, and the ways in which it reflects its datasets back into the lives of its users. Unlike many systems within the digital story‐ telling research space, GeoPoetry lacks the clear metrics of “narrative coherence” or “believable agents” to help assess whether it has successfully achieved its aesthetic goals. In this sense it is more like a work of digital media art, or electronic literature, where the criteria for success are more subjective and ephemeral. The kinds of questions we would like to ask of this system in any user study come down to its aesthetic and affective legibility: Do participants correctly perceive the sentiment expressed by the system? Does GeoPoetry have an impact on the experience of a participant, and how does that impact manifest itself? Does the system respond to events and changes in the geo-social landscape and is this response legible to listeners? There is also the bigger question about the suitability of Twitter as a dataset for this kind of location-based poetry. While it lends itself readily to a markov-chain text generation process, the data requires significant cleaning and filtering, and the underlying sentiments expressed on the platform tend to be predominantly negative. Future versions of GeoPoetry will explore other datasets, which will yield new challenges and insights into the design of locative media. We argue that GeoPoetry represents a new genre of generative storytelling systems that are situated in both time and space. While the aesthetic goals of our current system are oriented towards the production of a poetic abstraction of a geo-social context, there are many other approaches that could be taken using the same underlying infrastructure. GeoPoetry begins to open up the design space for real-time sonification of the invisible flows of data that surround travelers, and as more and more traces of our physical world are digitized and made available online the possibility space for authoring new forms of digital media art continues to expand.

References 1. Richter, H.: Dada: Art and Anti-Art. Thames & Hudson, New York (1997) 2. Friedman, K.: The Fluxus Reader. Wiley, Chicester (1998) 3. Sell, M.: Avant-garde Performance & the Limits of Criticism: Approaching the Living Theatre, Happenings/Fluxus, and the Black Arts Movement. University of Michigan Press, Ann Arbor (2008) 4. Knabb, K. (ed.): Situationist International Anthology. Bureau Of Public Secrets, Berkeley (2006) 5. Plant, S.: The Most Radical Gesture: The Situationist International in a Postmodern Age. Routledge, London (2002) 6. Sandford, M.: Happenings and Other Acts. Routledge, London (2003) 7. Boal, A.: Theater of the Oppressed. Pluto Press, London (2000) 8. Brecht, B.: Brecht on Theatre: The Development of an Aesthetic. Hill and Wang, London (1977) 9. Weiser, M.: The computer for the 21st century. Sci. Am. 265, 94–104 (1991) 10. Dourish, P., Bell, G.: Divining a Digital Future: Mess and Mythology in Ubiquitous Computing. The MIT Press, Cambridge (2011) 11. Montola, M., Stenros, J., Waern, A.: Pervasive Games: Theory and Design. Morgan Kaufmann Publishers Inc., San Francisco (2009)

GeoPoetry: Designing Location-Based Combinatorial Electronic Literature

95

12. Dey, A.K.: Understanding and Using Context. Pers. Ubiquit. Comput. 5, 4–7 (2001) 13. Schilit, B., Adams, N., Want, R.: Context-aware computing applications. In: First Workshop on Mobile Computing Systems and Applications, WMCSA 1994, pp. 85–90 (1994) 14. Dourish, P.: What we talk about when we talk about context. Pers. Ubiquit. Comput. 8, 19– 30 (2004) 15. Huizinga, J.: Homo Ludens. Taylor & Francis, New York (1949) 16. Salen, K., Zimmerman, E.: Rules of Play: Game Design Fundamentals. MIT Press, Cambridge, London (2004) 17. Lombardo, V., Damiano, R.: Storytelling on mobile devices for cultural heritage. New Rev. Hypermedia Multimed. 18, 11–35 (2012) 18. Wakkary, R., Hatala, M., Muise, K., Tanenbaum, K., Corness, G., Mohabbati, B., Budd, J.: Kurio: a museum guide for families. In: Proceedings of the 3rd International Conference on Tangible and Embedded Interaction, pp. 215–222. ACM, New York (2009) 19. Lu, M., Arikawa, M.: Map-Based Storytelling Tool for Real-World Walking Tour. In: Krisp, J.M. (ed.) Progress in Location-Based Services. LNGC, pp. 435–451. Springer, Heidelberg (2013) 20. Blythe, M., Reid, J., Wright, P., Geelhoed, E.: Interdisciplinary criticism: analysing the experience of riot! A location-sensitive digital narrative. Behav. Inf. Technol. 25, 127–139 (2006) 21. Reid, J., Geelhoed, E., Hull, R., Cater, K., Clayton, B.: Parallel worlds: immersion in locationbased experiences. In: CHI 2005, Extended Abstracts on Human Factors in Computing Systems, pp. 1733–1736. ACM, New York (2005) 22. Paay, J., Kjeldskov, J., Christensen, A., Ibsen, A., Jensen, D., Nielsen, G., Vutborg, R.: Location-based storytelling in the urban environment. In: Proceedings of the 20th Australasian Conference on Computer-Human Interaction: Designing for Habitus and Habitat, pp. 122– 129. ACM, New York (2008) 23. Nisi, V., Wood, A., Davenport, G., Oakley, I.: Hopstory: an interactive, location-based narrative distributed in space and time. In: Göbel, S., Spierling, U., Hoffmann, A., Iurgel, I., Schneider, O., Dechau, J., Feix, A. (eds.) TIDSE 2004. LNCS, vol. 3105, pp. 132–141. Springer, Heidelberg (2004) 24. Nisi, V., Oakley, I., Haahr, M.: Location-aware multimedia stories: turning spaces into places. In: Proceedings of the 4th International Conference on Digital Arts, pp. 72–82, Porto, Portugal (2008) 25. Crow, D., Pan, P., Kam, L., Davenport, G.: M-Views: a system for location-based storytelling. In: Proceedings of the 5th International Conference on Ubiquitous Computing. Association for Computing Machinery, Seattle, WA (2003) 26. Bichard, J., Brunnberg, L., Combetto, M., Gustafsson, A., Juhlin, O.: Backseat playgrounds: pervasive storytelling in vast location based games. In: Harper, R., Rauterberg, M., Combetto, M. (eds.) ICEC 2006. LNCS, vol. 4161, pp. 117–122. Springer, Heidelberg (2006). doi: 10.1007/11872320_14 27. Gustafsson, A., Bichard, J., Brunnberg, L., Juhlin, O., Combetto, M.: Believable environments: generating interactive storytelling in vast location-based pervasive games. In: Proceedings of the 2006 ACM SIGCHI International Conference on Advances in Computer Entertainment Technology. ACM, New York (2006) 28. Schöning, J., Hecht, B., Starosielski, N.: Evaluating automatically generated location-based stories for tourists. In: CHI 2008, Extended Abstracts on Human Factors in Computing Systems, pp. 2937–2942. ACM, New York (2008)

96

J. Rickman and J. Tanenbaum

29. Schöning, J., Hecht, B., Rohs, M., Starosielski, N., et al.: WikEar–Automatically generated location-based audio stories between public city maps. In: Adjun. Proceedings of Ubicomp 2007, pp. 128–131 (2007) 30. Procyk, J., Neustaedter, C.: GEMS: the design and evaluation of a location-based storytelling game. In: Proceedings of the 17th ACM Conference on Computer Supported Cooperative Work & Social Computing, pp. 1156–1166. ACM, New York (2014) 31. Russell, J.A.: A circumplex model of affect. J. Pers. Soc. Psychol. 39, 1161–1178 (1980) 32. Hutto, C.J., Gilbert, E.: Vader: a parsimonious rule-based model for sentiment analysis of social media text. In: Eighth International AAAI Conference on Weblogs and Social Media (2014) 33. Cambria, E., Olsher, D., Rajagopal, D.: SenticNet 3: a common and common-sense knowledge base for cognition-driven sentiment analysis. In: Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence, pp. 1515–1521. AAAI Press (2014) 34. Cambria, E., Livingstone, A., Hussain, A.: The hourglass of emotions. In: Esposito, A., Esposito, A.M., Vinciarelli, A., Hoffmann, R., Müller, V.C. (eds.) Cognitive Behavioural Systems. LNCS, vol. 7403, pp. 144–157. Springer, Berlin, Heidelberg (2012)