green_deer part 1
===


### General Questions ###


investigator: Okay, recording in progress. So, now we can continue to the next section. You should see "general questions regarding usage of digital libraries" now. 

green_deer: Yes.


##### GQ1 #####


investigator: Perfect. the first question here is which tasks do you usually use a digital library for? Please tick all answers, which apply and complete your own tasks.

Please give oral examples of the tasks you are ticking. So, for example, you could say, "okay, I tick "person search" because I like to keep track of myself".

green_deer: Okay. So, let's see, we have "person search", "find experts on topic". Yes. I guess that I'll check that box because that's exactly what I use it for. So, to keep track of experts in the topic also to look up myself because that's a nice way for having all of your research paper in one place, if you have a digital library, which aggregates all the information. 

Actually, the second box, I also check that one because yeah, search for latest papers, reviews and find paper on topics. "Venue search" not so much. Because yeah, it's seldom that in my community that a venue is a representative for a whole community. So, it's more spread out over multiple communities. 

"Get BibTeX data", definitely. That's like one of the most important use cases for it. "Get full text papers" well, we have access to most of, um yeah, to most of the vendors. So, I think, not sure if I would that ...

investigator: So, by vendors you mean, or what exactly? 

green_deer: I mean, like IEEE, ACM, stuff like that. 

investigator: Okay. But those are digital libraries.

green_deer: Indeed, they are. So, yeah. Yes, you're correct. So, let's check that box. "Study relations", study co-author relationships, study venue-author relationships, not so much. Okay. 

Okay. So, now for the first question...

investigator: Is there anything else any other tasks you usually do with a digital library, then you can include it under "Sonstiges".

green_deer: Mm, not that I know of right now. So, that's, so the questions before basically cover most of the use cases I used them for. 


##### GQ2 #####


investigator: Okay, perfect. Then we can continue with the next question. Which system or digital library do you usually use to solve these tasks? Please tick all answers, which apply and name others, which also apply, but are not given here. And please give a short oral description. Why you like, why use the system.

green_deer: Oh, okay. Let's start ACM of course. Because they have a ton of papers and a ton of venues, that's why I use the ACM digital library.

Bibsonomy, never used it. Same with Clarivate, same with dimensions. DBLP, I used that a lot because it has, I'd say, the most concise information aggregation I know of, from all the digital libraries, more or less, the same as ACM. So, I also check that. Google scholar is more like, I feel like Google scholar is very incomplete in terms of showing the papers or the related papers and also yeah, there are a lot of duplicate entries in Google Scholar I find, so I tick box, but I don't find the information that helpful. Semantic Scholar, I know it's there, but I also never use it. So, for Springer link, I'd say it's the same as ACM and Elsevier. So, I use it regularly. I totally ignore research gate and yeah. Usual search engines are basically the entry point for most of my searches. So, from Google, I get a link to ACM or, Springer or Elsevier or then go from there.

investigator: Okay. Is there any other system or digital library that you are using, which is not listed otherwise? 

green_deer: Not that I know of directly. No. 

investigator: Okay. Then we can continue with the next page, I think. So, you should now see task one.

green_deer: Yes. 


### TASK 1 ###


investigator: Okay. Task one, consider the following task, find two experts on a topic of your liking example topics could be "domain specific query languages" or "hashing functions", but should be from the broader or general area of computer and information science. So, you can pick any topic that you like, and you do not really have to do this task. We are only going to talk about how you would do this task usually.


##### TASK 1.1 #####


So, question one would be, what is your chosen topic? 

green_deer: Well, let's say "image recognition" or "image classification" if it's broad enough. 


##### TASK 1.2 #####


investigator: Yes, that's perfect. And how familiar are you with this topic?

green_deer: On a scale from one to 10? I'd say six to seven. 


##### TASK 1.3 #####


investigator: Okay, perfect. And next question is how would you define an expert?

green_deer: An author or co-author of a paper that really impacted or greatly impacted this community or this task over the recent years, there are a few papers that basically made tremendous impact and are basically used as the base, as the foundation for a lot of state of the art solutions.

investigator: Okay. Great. Do you want to add anything else to this definition or should we continue with the next question? 

green_deer: I think we can continue. 


##### TASK 1.4 #####


investigator: Okay, perfect. So, now how would you solve the task of finding two experts on a topic of your liking, of "image recognition"? 

green_deer: So, in general I would simply put exactly this search term into Google. So, "image recognition research papers", and this would lead me to probably the first two or three hits in this Google search would either guide me to Google scholar or to a blog post. And then from there, there's most likely a link to a digital library. To a paper, which is hopefully highly cited.

investigator: Okay. How do you check the citedness of the paper? 

green_deer: By looking at the H index or i10 index, I can do that via Google scholar, for instance. Also, there are probably other digital libraries that also provide this feature, but most of the time I'm opting for Google Scholar. 

investigator: Okay. So, now that you have the papers that are linked in like blog posts or something like that, and you have the H index of the people writing the paper and the citation score of the papers. How do you choose your experts and where?

green_deer: Well, most likely it's the case that if I have a set of papers, let's say candidate papers either we have a citation graph among them, so they cite each other. And then I find out which one was cited the most among those. and then I look at the author list of this yeah, of these papers that are cited the most in this candidate set of papers. This should be a representative of a highly cited paper inside the community and the author should thus, be an expert.

investigator: Okay. Where do you check the citation graph? 

green_deer: I'm not actually checking it. I'm reading the paper and check and looking at the citations they do and then see if there's any overlap between those papers. 

investigator: Okay. And for checking the author list of this highly cited paper, how do you go, what do you do there? Do you check the other names? Do you check the first author? Do you check the last author, which order and in which system do you look them up?

green_deer: In computer science it's there's a specific order in, on how people are listed on research papers. We have the first author, which does hopefully does bulk of the work.

And then the last author or the last authors are oftentimes, which is called the money positions. So, it's like the professors or the group leaders. So, it's those two ends of the author list that are important. 

investigator: Okay. And so now, how do you decide on your two experts?

green_deer: Well, it depends if the first author, which is oftentimes a PhD student or postdoc has multiple papers that are highly cited, then I consider this person the expert, but if there are multiple first authors associated to the same, yeah, professor on the money position, then I would consider the professor or the group to be the expert here. And the professor as the representative of that research group. 

investigator: Okay. Where do you look at the papers of the authors? 

green_deer: Whatever digital library I can find. I dunno if I understand the question correctly here. 

investigator: Yeah. it was just if you take the name and then do what to find the papers, so whatever digital library you like is okay. Do you want to add anything else to this task or do you think this describes your process perfectly?

green_deer: So, winding back to the citation graph I'd say, the way I construct this citation graph for me. So, why am I not doing this automatically? There are digital library features that give you this citation graph.

Uh, I do this manually because often it's important to see the citation in context. So, like in the related work of the paper, how do they cite it? Is it really the core or like the foundation or is it just yeah. Just the citation that needs to be there but doesn't bring anything to the table for the topic that's discussed in the paper.

It's like citing Donald Knuth in a hash map paper. You have to cite him because the foundation comes from him, but it's not relevant for newer works on the topic. 

investigator: Okay. So, do you want to add anything else to this task? 

green_deer: I think that's it. Thanks. 


### TASK 2 ###


investigator: Okay, perfect. So, then we can continue to the next page and the last task.

It's task two. Now, consider the following task, find relevant papers from a topic of your liking, which appeared after 2017. Example topics could be "paper recommendation" or "author disambiguation" but should be from a broader area of computer or information science. You can pick whatever topic you like, and you can also pick the topic from before, so "image recognition". 


##### TASK 2.1 #####


Now my first question would be what is your chosen topic? 

green_deer: So like a newer topic, I'd say. Let's go with "neural rendering", like NeRF or instant NGP. I hope that's broad enough. 

investigator: When did this topic come up?

green_deer: I'd say around 2020, maybe 19 or 20. 


##### TASK 2.2 #####


investigator: Okay, perfect. No, that's good. how familiar are you with this topic?

green_deer: I'd say four out of 10, maybe five. 


##### TASK 2.3 #####


investigator: Okay. And how would you define relevancy?

green_deer: It's actually pretty hard to, so it's a visual topic again. So, relevancy here is, does it improve the outcome? And as a human, you can basically just look at the outcome of the algorithm and decide if this is relevant or relevant improvement or not. So, that's. Yeah. I think that's how this whole visual community is driven.

So, yeah, I'd say it's basically looking at it. 


##### TASK 2.4 #####


investigator: Okay. Perfect. I think then we can go to question four. How would you solve the task of finding relevant papers from this topic, which appeared after 2017? Okay. That after 2017 part is now irrelevant, because the topic is newer than that. So, how would you find relevant papers from this topic?

green_deer: I'd say, so the problem here is that Google scholar and most other digital libraries lack behind this highly, yeah, rapidly advancing communities. So, most of the papers are preprinted to arXiv. So, that's one place to start and then probably also Twitter, because if there is a novel achievement in this community or a novel yeah, a novelty in this topic, then it's most likely put on Twitter so you can look at it and, yeah.

investigator: Okay. And how would you start searching for relevant papers then on arXiv or Twitter? 

green_deer: Probably arXiv. So, Twitter is more like, not a search engine for it. It pops up on your stream, but you can definitely search arXiv. 

investigator: Okay. And what would you search for and what would be your criterion be, to decide if it is relevant?

green_deer: it's a bit more difficult since there's not really a H index. so, it's really a yeah. By mouth propaganda. And I'd say because there isn't really, it's just a preprint. So, it's like how many people in the field agree on the preliminary results of that paper?

investigator: Okay. And this by mouth propaganda, where do you get this from? 

green_deer: Again, Twitter... 

investigator: Okay. So, what would be your entry point? Would your entry point be arXiv or would it be like a serendipity find on Twitter? 

green_deer: It depends if I'm actually searching or just, yeah. If I'm actively searching for this topic, I'd go to arXiv type in the topic name or examples of already existing implementations and see if I find relevant or associated paper with these keywords that are newer than the papers or the keywords that I put in. Like if I know instant NGP is from 2021 and I type in instant NGP and get a newer paper, which is from 2022, I consider it a candidate.

investigator: Okay. So, do you think this represents your process of finding relevant papers from this topic entirely, or is there anything else to add?

green_deer: I'd say it's hard. Again, it's hard to find relevant or structured information on such new and rapidly growing topics. So, it's the best I can find for now. The best workflow I can find for myself, yeah…

investigator: Okay, perfect. Then I think I will stop the recording, if you don't want to add anything else?

green_deer: That's it. Thank you very much. 


### Thank you ###


investigator: Okay. Thanks.

