Analysis and Visualization of Geo-Social Media, 12/07/2023, Alexander Dunkel

## Theoretic Background

<span class="smallgrey">Who's the audience? What are the goals?</span>

---

Audience: Landscape & Urban Planners

---

The "Public Perception" - Motivation

## EU Landscape Convention:

<strong>… landscape “as a zone or area as <span class="highlight">perceived</span> by local people or visitors” (ELC art. 1, para. 38). </strong>

Note:
- but it is very difficult to understand perception of many, 
  sometimes thousands of people

---

Problem(s):  
<strong>“Landscape” and perception (= valuation)  
of the landscape are inseparably intertwined.
</strong>

Note:
- illustration that tries to explain human-environment interaction

---

Notes:
- in practice, such theoretic concepts are often too complicated to implement  
- currently, 3 main sources for evaluating how people perceive environment

Let's ask an artificial intelligence to <span style="color:black">explain triangulation</span>..

---

Definition: <span class="highlight">Triangulation (Social Sciences)</span>

> By combining multiple observers, theories, methods, and empirical materials, 
researchers hope to overcome the weakness or intrinsic biases and the problems 
that come from single method, single-observer, and single-theory studies.

[Wikipedia](https://en.wikipedia.org/wiki/Triangulation_(social_science)),
see also Manson & O'Sullivan 2004 <sup><a href="https://doi.org/10.1068/a37100" class="smallsuper"><i data-customTooltip="Complexity theory in the study of space and place" class="fa fa-external-link"></i></a></sup>

Notes:
- Triangulation can have different meanings, please explain the most important ones.
- Triangulation is a term that is used in a variety of different fields, including psychology, sociology, and geometry. In psychology and sociology, triangulation is a method used to increase the validity of a research study by collecting data from multiple sources, using multiple methods, or both. This helps to ensure that the results are not influenced by any one particular perspective or method of data collection.

---

*Social Media data*:  
A supplementary source for planners to better understand  
(triangulate) <span style="color:#FF8D3E">Collective Human Behavior</span>

---

**Distinction**

|                      | <span class="highlight">VGI</span> | <span class="highlight">Geo-Social Media</span>                                             |
| -------------------- | ---------------------------------- | ------------------------------------------------------------ |
| Mode of contribution | active                             | passive                                                      |
| Topic                | A single, specific topic (or goal) | All of human environmental experience, recollection and imagination                        |
| Structure            | Pre-defined (relational) structure              | Differently structured (depending on network), not normalized, often Graph-based |
| Examples             | Wikipedia, OpenStreetMap, <i data-customTooltip="“to collect geographically representative photographs and information for every square kilometre of Great Britain and Ireland”">Geograph</i>           | Instagram, Flickr, Twitter                                   |
| License              | Specific License (e.g.: cc-by)     | Protected or no explicit license, sometimes mixed (Flickr)   |
| Application          | Experimental studies, clear goal               | Observational studies, Explorative                                        |

</span>

Ghermandi & Sinclair 2019 <sup><a href="https://doi.org/10.1016/j.gloenvcha.2019.02.003" class="smallsuper"><i data-customTooltip="Passive crowdsourcing of social media in environmental research: A systematic map" class="fa fa-external-link"></i></a></sup>

Task: Ask GPT-3 to explain the difference..

> Input: Explain the difference between Location Based Social Media Data (LBSM) and <span class="highlight">traditional Geographic Information (GI-)</span> Data.

---

<div style="position:absolute; margin-left: auto; margin-right: auto; margin-bottom: 100px; left: 0; bottom: 0;" class="fragment">      
<p style="text-align:left;color:#00FEFF">
4 Facets:<br>
Location (Where?)<br>  
User origin (Who?)<br>
Tags (What?)<br>
Time (When?)
</p> 
</div>
<video controls data-autoplay loop>
<source data-src="https://wwwpub.zih.tu-dresden.de/~s7398234/slide_assets/2017_FlickrEuropeAni.webm" type="video/webm" />
</video>

Notes:
- Visualisierung 72 M Fotos 2007-2017 in Europa (Flickr)
- Verteilung nicht zufällig: deutlich Gemeinsamkeiten der Wahrnehmung vieler Menschen erkennbar (z.B. Küstenlinien)
- 4 analytische Facetten der Daten

---

| <i data-customTooltip="Event Processing in Action">Etzion and Niblett (2010) </i>     | <span class="highlight"><i data-customTooltip="Prospects and challenges for social media data in conservation science">Minin et al. (2015)</i></span>  | <i data-customTooltip="An operational definition of context"> Zimmermann et al.(2007)</i> | <i data-customTooltip="A survey on context-aware systems">Dustdar and Rosenberg (2007)</i> | <span class="highlight"><i data-customTooltip="A conceptual framework for studying collective reactions to events in location-based social media">Dunkel et al. (2018)</i></span> |
| ----------------------------- | ------------------- | ----------------------- | --------------------------- | -------------------- |
| Temporal context              | When?               | Time                    | Time                        | Temporal Facet       |
| Spatial context               | Where?              | Situation               | Location                    | Spatial Facet        |
| Segmentation oriented context | Who?                | Role                    | Identity                    | Social Facet         |
| State-oriented context        | What?               | Interest Topic          | Activity/Status             | Thematic Facet       |

</span>

Different definitions of the <span class="highlight">four dimensions for characterizing behavior</span> (<u>highlighted</u>: social media specific)

<span class="fragment">Other systems: Chen, Lin & Yuan 2017 <sup><a href="https://doi.org/10.1111/cgf.13211" class="smallsuper"><i data-customTooltip="Social Media Visual Analytics" class="fa fa-external-link"></i></a></sup></span>

---

<span class="highgrey">4 Facets:</span>  
From Basic to complex Visualizations

WHERE (the Spatial Facet)

---

data:

<ul>
<li class="fragment" data-fragment-index="1"><span class="highlight">Explicit coordinates</span> (e.g. GPS via EXIF data, or photos "dropped" on a map <span class="highlight" data-customTooltip=">Example Flickr photo"><sup><a href="https://flickr.com/photos/64974314@N08/51617683785/in/datetaken-public/" class="smallsuper"><i class="fa fa-external-link"></i></a></sup></span>)</li>
<li class="fragment" data-fragment-index="2">Free-text descriptions of places (e.g. in posts) (<span class="highlight">Geoparsing <sup><a href="https://www.southampton.ac.uk/~sem03/SIGIR-2019-poster-v2.pdf" class="smallsuper"><i data-customTooltip=">Overview by Stuart et al. (2019)" class="fa fa-external-link"></i></a></sup></span>)</li>
<li class="fragment" data-fragment-index="3">Language of a post (<span class="highlight">Language detection</span>)</li>
<li class="fragment" data-fragment-index="4">Based on past user behavior (e.g. <span class="highlight">Bayesian Inference <sup><a href="https://www.researchgate.net/publication/326192551_Location_Inference_for_Non-Geotagged_Tweets_in_User_Timelines" class="smallsuper"><i data-customTooltip=">Li et al. (2018)" class="fa fa-external-link"></i></a></sup></span>)</li>
<li class="fragment" data-fragment-index="5"><span class="highlight">Gazetteer</span>: A database of places (e.g. Geonames <sup><a href="https://www.geonames.org/" class="smallsuper fa fa-external-link"></a></sup>; or "Knäbäckshusens beach, Sweden" on Instagram <sup><a href="https://www.instagram.com/explore/locations/270772352/knabackshusens-underbara-bad/" class="smallsuper fa fa-external-link"></a></sup></li>
<li class="fragment" data-fragment-index="5">...</li>
</ul>

</div>

Notes:
- toponyms

---

<div style="height:99vh">
<img style="width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_where.svg" />
<br><br>
methods of processing:
<br><br>

<ul>
<li class="fragment" data-fragment-index="1">"object" related approaches:</li>
    <ul class="fragment" data-fragment-index="1">
    <li>clustering (with influence of cluster distance)</li>
    <li>Voronoi diagrams</li>
    </ul>
<li class="fragment" data-fragment-index="2">"space" related (grid based approaches):</li>
    <ul>
    <li class="fragment" data-fragment-index="2">binary search</li>
    <li class="fragment" data-fragment-index="3">geohash</li>
    <li class="fragment" data-fragment-index="4">quadtree (e.g. based on density criteria)</li>
    </ul>
</ul>

<p class="fragment" data-fragment-index="5">Bereuter 2015, p.27 <sup><a href="https://www.geo.uzh.ch/dam/jcr:2d048771-fb28-4612-a228-c7628d74be51/Bereuter-2015-PhD-Thesis.pdf#page=45" class="smallsuper"><i data-customTooltip="Quadtree-based Real-time Point Generalisation for Web and Mobile Mapping" class="fa fa-external-link"></i></a></sup></p>

</div>

Notes:
- ask for clustering methods:
- k-means, mean shift, HDBSCAN

---

<div style="width:100%;position:absolute;top:0px;text-align:center;z-index:9998;">
<p class="highlight" style="vertical-align: top;background-color:rgb(255,255,255,0.8)" >
<strong>HDBSCAN Photo Location Clustering at <br> Highpark, Toronto</strong> 
</p> 
</div>
</div>

Note:
Data (photo location) groups itself along similar patterns of density, beginning with the highest density areas
bottom-up clustering: joining things that belong together
question: what is the best clustering distance?

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_where.svg" />
<br>
binary search

Zacharatou et al. 2021 <sup><a href="https://www.researchgate.net/publication/348334913_The_Case_for_Distance-Bounded_Spatial_Approximations" class="smallsuper"><i data-customTooltip="The Case for Distance-Bounded Spatial Approximations" class="fa fa-external-link"></i></a></sup>

</div>

---

Parquet Format (uses binary search): <br>Visualization of 350 Million Geo-Social Media posts from Twitter <sup><a href="https://ad.vgiscience.org/twitter-global-preview/00_Twitter_datashader.html" class="smallsuper fa fa-external-link"></a></sup>

</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_where.svg" />
<br><br><br><br>
Geohash:

- "Snapping" points to a grid
    - coordinate: <span class="highlight">`51.0281,13.7267` = Geohash (7) `u31f239`</span> <sup><a href="https://www.movable-type.co.uk/scripts/geohash.html?geohash=u31f239" class="smallsuper fa fa-external-link"></a></sup>
- granularity of the grid is based on length of geohash-string

</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_where.svg" />
<br>
Quadtree:

<p class="small fragment" data-fragment-index="1" style="position:absolute;right:150px;top:430px;text-align:left">
A friendly <br>user on Stackoverflow <sup><a href="https://stackoverflow.com/a/48355534/4556479" class="smallsuper fa fa-external-link"></a></sup>
</p>

<div style="position:relative;top:450px;">
<ul>
<li>Adaptive, recursive grid division based on some criteria</li>
<ul class="fragment highlight" data-fragment-index="1"><li>e.g. <code>n ≤ 2</code></li></ul>
</ul>
</div>
</div>

---

<ul>
<li class="fragment" data-fragment-index="1">visualizations of locations in maps, e.g.</li>
    <ul>
    <li class="fragment" data-fragment-index="2">static or animated points (from where people tweet)</li>
    <li class="fragment" data-fragment-index="3">points aggregated to areas (grid cells, NUTS, …), or 3D-Bars (etc.)</li>
    <li class="fragment" data-fragment-index="4">connected points (edges and nodes), e.g. Eric Fisher's "Tourists & Locals"<sup><a href="https://www.flickr.com/photos/walkingsf/albums/72157624209158632/" class="smallsuper fa fa-external-link"></a></sup></li>
    </ul>
<li class="fragment" data-fragment-index="5">heatmaps – vernacular regions / vague places (Hollenstein and Purves, 2010 <sup><a href="https://doi.org/10.5311/JOSIS.2010.1.3" class="smallsuper"><i data-customTooltip="Exploring place through user-generated content: Using Flickr to describe city cores" class="fa fa-external-link"></i></a></sup>)</li>
</ul>
</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_where.svg" />
<br>
<div style="z-index:9998;width:100%;position:relative;background-color:white">
<span class="highlight">Binary search (grid based): <br>Global reactions to Sunset & Sunrise <sup><a href="https://anonymous-peer12345.github.io/resources/html/01_grid_agg.html#Prepare-binary-search" class="smallsuper fa fa-external-link"></a></sup></li></span>
</div>

</div>

WHAT (the Topical Facet)

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_what.svg" />
<br><br><br>
data:

<ul>
<li class="fragment" data-fragment-index="1"><span class="highlight">(hash-) tags and emoji</span></li>
<li class="fragment" data-fragment-index="2"><span class="highlight">the textual content</span> (e.g. of tweets, or the descriptions of photos and photo titles on Flickr)</li>
<li class="fragment" data-fragment-index="3">the content of the <span class="highlight">photo/video itself</span>, e.g. automatic image classification, labeling</li>
<li class="fragment" data-fragment-index="4">Other <span class="highlight">contextual information</span> (e.g. post type: is it a retweet? a comment? a "cross-post" from Instagram? or original content?)</li>
<li class="fragment" data-fragment-index="5">...</li>
</ul>
</div>

---

<ul>
<li class="fragment" data-fragment-index="1"><span class="highlight">basic filtering</span> (blocklists, allowlists <sup><a href="https://ad.vgiscience.org/meingruen/meingruen_activities_ipynb.html#Topic-Selection-and-Ranks" class="smallsuper"><i data-customTooltip="Actvity Selection for Comparison ATKIS LBSM-DE" class="fa fa-external-link"></i></a></sup>), <br>counting & sorting</li>
<li class="fragment" data-fragment-index="2">Natural-language Processing (<span class="highlight">NLP</span>):</li>
    <ul>
    <li class="fragment" data-fragment-index="3">Term-Frequency Inverse Document Frequency (<span class="highlight">TF-IDF</span>)</li>
    <li class="fragment" data-fragment-index="4"><span class="highlight">Cosine Similarity</span> (e.g. see Wartmann & Purves 2018 <sup><a href="https://doi.org/10.1016/j.landurbplan.2018.03.021" class="smallsuper"><i data-customTooltip="Investigating sense of place as a cultural ecosystem service in different landscapes through the lens of language" class="fa fa-external-link"></i></a></sup>)</li>
    </ul>
</ul>

</div>

Notes:
- "Explain in simple words Term-Frequency Inverse Document Frequency (TF-IDF)."
- "Explain in simple words Cosine Similarity and what word vector means in this context. Can you illustrate example words and their vector similarity?"

---

methods of processing:

<ul>
<li><span class="highlight">basic filtering</span> (blocklists, allowlists <sup><a href="https://ad.vgiscience.org/meingruen/meingruen_activities_ipynb.html#Topic-Selection-and-Ranks" class="smallsuper"><i data-customTooltip="Actvity Selection for Comparison ATKIS LBSM-DE" class="fa fa-external-link"></i></a></sup>), <br>counting & sorting</li>
<li>Natural-language Processing (<span class="highlight">NLP</span>):</li>
    <ul>
    <li>Umbrella term for a number of tasks for <span class="highlight">processing human language as it is spoken or written</span> </li>
    <li>Term-Frequency Inverse Document Frequency (<span class="highlight">TF-IDF</span>)</li>
    <li><span class="highlight">Cosine Similarity</span> (e.g. see Wartmann & Purves 2018 <sup><a href="https://doi.org/10.1016/j.landurbplan.2018.03.021" class="smallsuper"><i data-customTooltip="Investigating sense of place as a cultural ecosystem service in different landscapes through the lens of language" class="fa fa-external-link"></i></a></sup>)</li>
    </ul>
</ul>

</div>

---

methods of processing:

<ul>
<li><span class="highlight">basic filtering</span> (blocklists, allowlists <sup><a href="https://ad.vgiscience.org/meingruen/meingruen_activities_ipynb.html#Topic-Selection-and-Ranks" class="smallsuper"><i data-customTooltip="Actvity Selection for Comparison ATKIS LBSM-DE" class="fa fa-external-link"></i></a></sup>), <br>counting & sorting</li>
<li>Natural-language Processing (<span class="highlight">NLP</span>):</li>
    <ul>
    <li>Umbrella term for a number of tasks for <span class="highlight">processing human language as it is spoken or written</span> </li>
    <li>challenges (e.g.): meaning & homonyms, filling words, sentiment, sarcasm, metaphor, irony, attitudes, confusion etc.</li>
    <li>Term-Frequency Inverse Document Frequency (<span class="highlight">TF-IDF</span>)</li>
    <li><span class="highlight">Cosine Similarity</span> (e.g. see Wartmann & Purves 2018 <sup><a href="https://doi.org/10.1016/j.landurbplan.2018.03.021" class="smallsuper"><i data-customTooltip="Investigating sense of place as a cultural ecosystem service in different landscapes through the lens of language" class="fa fa-external-link"></i></a></sup>)</li>
    </ul>
</ul>

</div>

Notes:
- Words can have different meanings depending on the context (homonyms)

---

> semantic similarity between two documents

<span class="fragment">
<ul><span class="highlight">Documents</span> could be:
<li>Wikipedia articles</li>
<li>Social Media posts</li>
<li>Movies</li>
<li>etc.</li>
</ul>
</span>
</div>

---

> Cosine similarity is the cosine of the angle between the vectors; that is, it is the dot product of the vectors divided by the product of their lengths.

Wikipedia <sup><a href="https://en.wikipedia.org/wiki/Cosine_similarity" class="smallsuper fa fa-external-link"></a></sup>
</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_what.svg" />
<br><br><br><br><br>
Example: Similarity of "<span class="highlight">Barock</span>", "<span class="highlight">Dresden</span>", "<span class="highlight">University</span>"

```
Dresden ==>>    [1,0,0]
Barock ==>>     [0,1,0]
University ==>> [0,0,1]
```

</div>

Notes:
- assign to a 3 dimensional vector space

---

Word count frequency for all documents ("corpus"):

| Documents  | <span class="highlight">"Dresden"</span> | <span class="highlight">"Barock"</span> | <span class="highlight">"University"</span> |
| ------------ | ------- | ------ | ---------- |
| TUD <sup><a href="https://de.wikipedia.org/wiki/Frauenkirche_(Dresden)" class="smallsuper fa fa-external-link"></a></sup>          |  373       |   0     |   191         |
| Zwinger <sup><a href="https://de.wikipedia.org/wiki/Zwinger_(Dresden)" class="smallsuper fa fa-external-link"></a></sup>     |   137      |  31      |     1       |
| Frauenkirche <sup><a href="https://de.wikipedia.org/wiki/Frauenkirche_(Dresden)" class="smallsuper fa fa-external-link"></a></sup> |  124       |  26      |    1        |

</span>

```
corpus_vec = [[373 0 191]
              [137 31  1]
              [124 26  1]]
```

</span>
</div>

---

$$ cos(\alpha) = 0,41 $$

</span>

<p class="fragment" style="position:absolute;right:0px;top:550px;text-align:left">
see Pippig 2014 <sup><a href="https://tud.qucosa.de/api/qucosa%3A28589/attachment/ATT-0/" class="smallsuper fa fa-external-link"></a></sup>
</p>

</div>

Notes:
- Smaller angles between vectors produce larger cosine values, 
- large values indicate greater cosine similarity.

---

Cosine Similarity Issues:
- In a given document we shouldn’t count repeating words too much.
- There are words that are just very common, so they appear in lots of documents. (“the”, “and”, “or” etc..)
- Syntactic structure of sentences is lost.

</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_what.svg" />
<br>
<div style="z-index:9998;width:100%;position:relative;background-color:white">
<span class="highlight">TF-IDF & Cosine Similarity (Uganda)<br>for reactions to sunset and sunrise <sup><a href="https://anonymous-peer12345.github.io/resources/html/06_semantics.html" class="smallsuper fa fa-external-link"></a></sup></li></span>
</div>

</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_what.svg" />
<br>
<div style="z-index:9998;width:100%;position:relative;background-color:white">
<span class="highlight">TF-IDF & Cosine Similarity (Indonesia)<br>for reactions to sunset and sunrise <sup><a href="https://anonymous-peer12345.github.io/resources/html/06_semantics.html" class="smallsuper fa fa-external-link"></a></sup></li></span>
</div>
<div style="width:100%;height:100%">
<iframe style="width:100%;height:100%;position:relative;top:-50px" frameBorder="0" data-src="assets/sunset_cosine_flickr_IDN.html"></iframe>
</div>

</div>

---

visualization methods:

- word clouds, georeferenced tag maps
- thematic filtered maps

</div>

---

<img style="height:500px" class="plain"  src="images/topic_classification.png"/>
<span class="fragment">
<code>['event','music','festival','concert']</code> <br>Gugulica 2019 <sup><a href="https://ad.vgiscience.org/mobile_cart_workshop2020/04_topic_classification.html" class="smallsuper fa fa-external-link"></a></sup>
</span>
</div>

---

Yang, MacEachren, & Yang 2019 <sup><a href="https://doi.org/10.1080/00087041.2018.1533294" class="smallsuper fa fa-external-link"></a></sup>

</div>

WHO (the Social Facet)

---

data:

<ul>
<li class="fragment" data-fragment-index="1"><span class="highlight">profile data</span></li>
    <ul>
    <li class="fragment" data-fragment-index="1">e.g. the public "Home Location" on Flickr <br>user profiles: Text-based description of a location (<span class="highlight" data-customTooltip=">Example Flickr user Profile (me) - with false information (!)">Geocoding <sup><a href="https://www.flickr.com/people/64974314@N08/" class="smallsuper"><i class="fa fa-external-link"></i></a></sup></span>)</li>
    </ul>
<li class="fragment" data-fragment-index="3">Language of a post (<span class="highlight">Language detection</span>)</li>
<li class="fragment" data-fragment-index="3">Gender, Age (e.g. <span class="highlight">detection through Deep learning </span> <sup><a href="https://link.springer.com/chapter/10.1007/978-3-030-63128-4_26" class="smallsuper"><i data-customTooltip=">Kowsari et al. (2020)" class="fa fa-external-link"></i></a></sup></span>)</li>
<li class="fragment" data-fragment-index="4">past user behavior (e.g. <span class="highlight">Bayesian Inference <sup><a href="https://www.researchgate.net/publication/326192551_Location_Inference_for_Non-Geotagged_Tweets_in_User_Timelines" class="smallsuper"><i data-customTooltip=">Li et al. (2018)" class="fa fa-external-link"></i></a></sup></span>)</li>
<li class="fragment" data-fragment-index="5">User classification based on <span class="highlight">network connections</span> (follows, friends, comments, likes, retweets, cross-network profile links)</li>
<li class="fragment" data-fragment-index="5">...</li>
</ul>

</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_who.svg" />
<br>
methods of processing:

<ul>
<li class="fragment"><span class="highlight">taxonomy</span> for classifying user groups <br>(Gründemann & Burghadt 2020 <sup><a href="https://doi.org/10.5194/agile-giss-1-5-2020" class="smallsuper"><i data-customTooltip="A taxonomy for classifying user groups in location-based social media" class="fa fa-external-link"></i></a></sup></li>
<li class="fragment"><span class="highlight">user profiling</span> based on preferences / user characteristics (Cambridge analytica)</li>
    <ul><li class="fragment">e.g. the "Big Five" personality traits: <sup><a href="https://bigthink.com/strange-maps/geopsychology-regional-personality-variation/" class="smallsuper"><i data-customTooltip="The Big Five personality traits, a.k.a. the CANOE model. There is also a geographic component to their distribution, research suggests." class="fa fa-external-link"></i></a></sup></li>
        <ul><li class="fragment">openness  (curious vs. cautious)</li>
        <li class="fragment">conscientiousness (organized vs. careless)</li>
        <li class="fragment">extraversion (outgoing vs. reserved)</li>
        <li class="fragment">agreeableness (compassionate vs. rational)</li>
        <li class="fragment">neuroticism (sensitive vs. resilient)</li>
    </ul>
</ul>

</div>

---

- force-directed graphs
- diagrams, summary schemes, distribution visualizations (bar graphs etc.)
- maps
- ...

</div>

---

<p class="small" style="position:absolute;right:-50px;top:430px;text-align:left">
Force-directed graph:<br>
Twitter network relationship <br>of members of the German parliament <br>Gründemann & Burghadt 2020 <sup><a href="https://doi.org/10.5194/agile-giss-1-5-2020" class="smallsuper"><i data-customTooltip="A taxonomy for classifying user groups in location-based social media" class="fa fa-external-link"></i></a></sup>
</p>

</div>

---

[<img class="plain"  src="images/07_repr_map_glob_americans.png"/>](https://flic.kr/p/DxoZjY)  
<span style="color:#FF8D3E">Americans</span>

</div>

---

[<img class="plain"  src="images/07_repr_map_glob_europeans.png"/>](https://flic.kr/p/E417S5)  
<span style="color:#FF8D3E">Europeans</span>

</div>

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_who.svg" />
<br><br>
<a href="http://blog.alexanderdunkel.com/2016/02/differing-photo-patterns-based-on-user-origin/"><img class="plain"  src="images/08_jacobsweg.png"/></a> 
<br><strong>a:</strong> All Photo Locations  
<br><span class="fragment">    <strong>b:</strong> Photo Locations from Germans</span>  
<br><span class="fragment highlight"><strong>➝</strong> Camino de Santiago</span>

</div>

WHEN (the Temporal Facet)

---

<div style="height:99vh">
<img style="z-index:9999;width:300px;height:auto;transform: rotate(45deg);position:absolute;top:100px;right:0px" src="images/banner_when.svg" />
<br>
data:

<ul>
<li class="fragment">Photo timestamps (<span class="highlight">Creation time</span>)</li>
<li class="fragment"><span class="highlight">Upload time</span> (e.g. Instagram)</li>
<li class="fragment">GPS Time Stamps, GPS Date Stamps <br>(e.g. EXIF data on Flickr)</li>
<li class="fragment"><span class="highlight">inferred</span> based on other content (e.g. night, day, winter, summer detected from pictures)</li>
<li class="fragment"><span class="highlight">based on references</span> to specific events (e.g. use of "sunset" or "sunrise" terms in text; or <code>#fullforcefestival2019</code>)</li>
</ul>

</div>

---

- streaming data
- information spread analysis
- comparison graphics (e.g. before and after an event)
- event detection/ threshold analysis

</div>

---

<ul>
<li class="fragment"><span class="highlight">frequency diagrams</span> over time with different granularity / periodical</li>
<li class="fragment"><span class="highlight">monitoring</span> tools</li>
<li class="fragment"><span class="highlight">animations</span></li>
</ul>

</div>

---

<p class="small" style="z-index:9999;position:absolute;right:0px;top:450px;text-align:left">
circular diagram <br>with bar charts <br>(Dunkel 2015 <sup><a href="https://doi.org/10.1016/j.landurbplan.2015.02.022" class="smallsuper"><i data-customTooltip="Visualizing the perceived environment using crowdsourced photo geodata" class="fa fa-external-link"></i></a>)</sup></p>

</div>

---

<p style="z-index:9999;width:100%;position:relative;top:-150px;"><span class="highlight">Global Spread of Information</span></p>

<div style="width:550px;position:absolute;top:400px;right:0px;text-align:right;z-index:9999;background-color:white">
<em><p style="vertical-align: bottom" class="small">
Frequentation of popular </br>vantage points in Europe </br>from Instagram data </br>Claudia Tautenhahn, </br>Master thesis WS 2019/20
</p></em>
</div>
</div>

Notes:
- periodicity

---

**Monitoring**

</div>

---

</div>

---

[<img class="plain"  src="images/11_monitor.png"/>](http://maps.alexanderdunkel.com/?mapid=2)

</div>

---

<div style="width:100%;position:absolute;top:50px;text-align:center;z-index:9999;">
<p class="highlight" style="vertical-align: top;" >
<strong>Temporal patterns of Flickr photo <br>contribution in California, 2009-2011</strong> 
</p> 
</div>
</div>

Bias and normalization

<a href="https://chat.openai.com/chat">AI</a> explain?

> Phrase a question!

> What are types of bias in Volunteered Geographic Information (VGI) and User-generated Content (UGC) and why is it important to normalize data when visualizing such data? Name some common quantity measurements that are used to visualize VGI and UGC data.

---

> A bias is a systematic error impairing results or the validity of a study.

Delgado-Rodríguez & Llorca 2004 <sup><a href="https://doi.org/10.1136/jech.2003.008466" class="smallsuper fa fa-external-link"></a></sup>

---

Common types of bias and how to avoid them:

<ul>
<li><span class="highlight">data sampling or selection bias</span>, e.g. some users dominating the discourse</li>
    <ul>
    <li class="fragment" data-fragment-index="1">Use of <span class="highlight">explicit metrics and quantity measurements</span>, e.g. „usercount“, „postcount“, „userdays“ </li>
    <li class="fragment" data-fragment-index="2"><span class="highlight">Highlighting the limitations</span> of a study, since results are usually not <em>universally</em> representative (e.g. the digital divide)</li>
    </ul>
</ul>

---

<img class="plain" style="width:500px" src="images/userdays.png"/><br>
<span style="color:#FF8D3E">"Sunrise" userdays in Europe</span>, Jupyter notebook <sup><a href="https://anonymous-peer12345.github.io/resources/html/01_grid_agg.html" class="smallsuper fa fa-external-link"></a></sup>

---

<img class="plain" style="width:500px" src="images/usercount.png"/><br>
<span style="color:#FF8D3E">"Sunrise" usercount in Europe</span>

---

<ul>
<li><span class="highlight">processing bias or algorithmic bias</span></li>
    <ul>
    <li class="fragment" data-fragment-index="1">explicit preparation of <span class="highlight">input parameters</span> that affect processing, instead of hard-coded thresholds that are non-transparent to the user</li>
    <li class="fragment" data-fragment-index="2"><span class="highlight">Avoiding</span> the use of <span class="highlight">absolute frequencies</span>, e.g. through the use of statistical methods such as TF-IDF, cosine similarity, or chi-square </li>
    </ul>
</ul>

---

<ul>
<li><span class="highlight">researcher bias</span>, e.g. prejudicial or close-minded interpretation</li>
    <ul>
    <li class="fragment" data-fragment-index="1"><span class="highlight">triangulation</span> using different data sources and processing methods (external validity testing)</li>
    <li class="fragment" data-fragment-index="2">testing of results with <span class="highlight">different parameters or comparing results across different slices of data</span> (internal consistency checks)</li>
    </ul>
</ul>

---

<ul>
<li><span class="highlight">mapping/visualization bias</span></li>
    <ul>
    <li class="fragment" data-fragment-index="1">visualization of relative quantities instead of absolute counts, e.g. <span class="highlight">signed chi-square</span> (Clarke et al. 2007 <sup><a href="https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=4376138" class="smallsuper"><i data-customTooltip="Interactive Visual Exploration of a Large Spatio-Temporal Dataset: Reflections on a Geovisualization Mashup" class="fa fa-external-link"></i></a></sup>)</li>
    <li class="fragment" data-fragment-index="2"><span class="highlight">head / tail breaks</span> as a classification algorithm for data with heavy-tailed distributions (Zipf´s law) (Jiang 2013 <sup><a href="https://arxiv.org/ftp/arxiv/papers/1209/1209.2801.pdf" class="smallsuper fa fa-external-link"></a></sup>)</li>
    <li class="fragment" data-fragment-index="3">Avoiding of overlapping symbols (e.g. point visualizations), or highlighting non-printed elements</li>
    </ul>
</ul>

..more complex (real-world) application examples

---

<img class="plain" style="height:500px;" src="images/scatterblogs.png"/>
<br>
Scatterblog "<span class="highlight">Dashboard</span>": <code>#Hochwasser</code> - Using Visual Analytics of social media in civil protection

<p class="small" style="position:absolute;right:-50px;top:400px;text-align:left">
Zisgen, Kern, <br>Thom, Ertl 2014 <sup><a href="https://www.researchgate.net/publication/261633257_Hochwasser_-_Visuelle_Analyse_von_Social_Media_im_Bevolkerungsschutz" class="smallsuper fa fa-external-link"></a></sup>
</p>

---

(composite facet "where" and "what")

---

<div style="width:100%;position:absolute;bottom:-700px;text-align:center;z-index:9999;" class="fragment" data-fragment-index="1">
<p style="vertical-align: bottom;background-color:rgba(255, 255, 255, 0.8);" >
<strong>Tag Map for Fort Mason Center, San Francisco:</strong> 
Number of photo locations: <span style="color:#FF8D3E">41,777</span> Contributing users: <span style="color:#FF8D3E">5002</span> Total available tags: <span style="color:#FF8D3E">197,858</span> Period (date taken): <span style="color:#FF8D3E">2007-2014</span>
</p> 
</div>

Note:
Fort Mason Design Competition 2012 (Team West 8)  
- engaging with perception of others means showing empathy 
- crucial when designing public spaces

---

[<img class="plain"  src="images/03_scales.png"/>](https://ad.vgiscience.org/tagmaps/docs/theory/introduction/)  
<div style="width:100%;position:absolute;bottom:0px;text-align:center;z-index:9999;">
<p style="vertical-align: bottom;" >
Tag Maps at different scales (Greater Toronto Area)
</p> 
</div>

Please vote: E-Learning-Schmuckstück 2023<br>  
<span class="fragment"><strong><a href="https://tud.link/wud6">https://tud.link/<span class="highlight">wud6</span></a></strong></span>