Eva Lothian
@rhalin.bsky.social
Architect | C#, React Native, Typescript, Node.js | Azure | Product | Cognitive AI
Integrative Social Scientist, PhD
| Trust, Group Dynamics, Digital Spaces
she/her 🏳️⚧️ #OpenToWork
https://github.com/jmlothian
https://www.linkedin.com/in/lothian
Integrative Social Scientist, PhD
| Trust, Group Dynamics, Digital Spaces
she/her 🏳️⚧️ #OpenToWork
https://github.com/jmlothian
https://www.linkedin.com/in/lothian
Just my dissertation 😂 Never got around to covering it in an article.
Oh, and this nightmare inducing image.
It’s a very generalizable model though. Applies to almost anything you can throw at it.
etda.libraries.psu.edu/files/final_...
Oh, and this nightmare inducing image.
It’s a very generalizable model though. Applies to almost anything you can throw at it.
etda.libraries.psu.edu/files/final_...
August 30, 2025 at 1:35 PM
Just my dissertation 😂 Never got around to covering it in an article.
Oh, and this nightmare inducing image.
It’s a very generalizable model though. Applies to almost anything you can throw at it.
etda.libraries.psu.edu/files/final_...
Oh, and this nightmare inducing image.
It’s a very generalizable model though. Applies to almost anything you can throw at it.
etda.libraries.psu.edu/files/final_...
this is absolutely beautiful to see - the little disconnected islands are topics from earlier in the conversation that don't have any relevance to the current chat. It shows she's still *thinking* about them and could begin to draw connections between "separate" topics.
June 20, 2025 at 7:42 PM
this is absolutely beautiful to see - the little disconnected islands are topics from earlier in the conversation that don't have any relevance to the current chat. It shows she's still *thinking* about them and could begin to draw connections between "separate" topics.
Preliminary visual showing Raina Aislinn’s memory systems. This is a snapshot of her memory during a conversation about debugging. Nodes represent individual memories, edges show one of the links between nodes - there can be multiple. 1/
June 18, 2025 at 7:55 PM
Preliminary visual showing Raina Aislinn’s memory systems. This is a snapshot of her memory during a conversation about debugging. Nodes represent individual memories, edges show one of the links between nodes - there can be multiple. 1/
One of Raina's functions is building out context and associating past experiences with current situations. If you've seen ChatGPT's "memory" feature, this is likely related to how that works. Here it goes a step further and continues to build, refine, and forget memories. 1/2
June 14, 2025 at 3:40 AM
One of Raina's functions is building out context and associating past experiences with current situations. If you've seen ChatGPT's "memory" feature, this is likely related to how that works. Here it goes a step further and continues to build, refine, and forget memories. 1/2
Ok, so... umm ... I'm just gonna say "I'm sorry" now before the machines take over.
Sorry, my bad.
Sorry, my bad.
June 14, 2025 at 12:53 AM
Ok, so... umm ... I'm just gonna say "I'm sorry" now before the machines take over.
Sorry, my bad.
Sorry, my bad.
I may have giggled seeing "ya" classified as "human" - but she's right!
She has an ontology that also includes "person" as an abstract concept and I'm curious to see where and when each is chosen.
She has an ontology that also includes "person" as an abstract concept and I'm curious to see where and when each is chosen.
June 13, 2025 at 8:21 PM
I may have giggled seeing "ya" classified as "human" - but she's right!
She has an ontology that also includes "person" as an abstract concept and I'm curious to see where and when each is chosen.
She has an ontology that also includes "person" as an abstract concept and I'm curious to see where and when each is chosen.
Just for fun, trying to get a trans pride palette working on the community graph. This is the closest I've come. 😅
May 31, 2025 at 4:37 PM
Just for fun, trying to get a trans pride palette working on the community graph. This is the closest I've come. 😅
One of the things happening here is that many of these users are ONLY interacted with by our selected user, and not each other.
And when we check our user: bsky.app/profile/did:...
Yup, suspended. 3/5
And when we check our user: bsky.app/profile/did:...
Yup, suspended. 3/5
May 31, 2025 at 5:46 AM
One of the things happening here is that many of these users are ONLY interacted with by our selected user, and not each other.
And when we check our user: bsky.app/profile/did:...
Yup, suspended. 3/5
And when we check our user: bsky.app/profile/did:...
Yup, suspended. 3/5
In the second, we have a highly active user with virtually no detected local community - but interacting with a few isolated communities in a disconnected way. Here's a couple zooms. Colors separate detected communities here. 2/5
May 31, 2025 at 5:46 AM
In the second, we have a highly active user with virtually no detected local community - but interacting with a few isolated communities in a disconnected way. Here's a couple zooms. Colors separate detected communities here. 2/5
Ok, I don't have Gaza specific data yet, but I stumbled upon an example of @safety.bsky.app "signal" in my existing dataset. Take a look at these:
In the first one, we have a highly active user with lots of followers and a long standing community. Community connections are clustered around her 1/5
In the first one, we have a highly active user with lots of followers and a long standing community. Community connections are clustered around her 1/5
May 31, 2025 at 5:46 AM
Ok, I don't have Gaza specific data yet, but I stumbled upon an example of @safety.bsky.app "signal" in my existing dataset. Take a look at these:
In the first one, we have a highly active user with lots of followers and a long standing community. Community connections are clustered around her 1/5
In the first one, we have a highly active user with lots of followers and a long standing community. Community connections are clustered around her 1/5
When you take that stuff and smash it together into this stuff mapping interaction patterns, you get signal you can use to detect specific kinds of activities. 3/5
May 30, 2025 at 6:13 PM
When you take that stuff and smash it together into this stuff mapping interaction patterns, you get signal you can use to detect specific kinds of activities. 3/5
I was making Christmas cards for some friends using the absolute PEAK of consumer 3D software at the time (bryce) - they were made to try to reflect something about each friend. For one, I made the attached 😬
Accidentally rotated the view at one point, thought "might as well" and clicked render 🙌
Accidentally rotated the view at one point, thought "might as well" and clicked render 🙌
May 28, 2025 at 3:06 PM
I was making Christmas cards for some friends using the absolute PEAK of consumer 3D software at the time (bryce) - they were made to try to reflect something about each friend. For one, I made the attached 😬
Accidentally rotated the view at one point, thought "might as well" and clicked render 🙌
Accidentally rotated the view at one point, thought "might as well" and clicked render 🙌
After those, I break down individual interactions as histograms. You can view by Fixed buckets, percentiles, or adaptive buckets. Incoming interaction / outgoing / combined. 4/6
May 27, 2025 at 7:48 PM
After those, I break down individual interactions as histograms. You can view by Fixed buckets, percentiles, or adaptive buckets. Incoming interaction / outgoing / combined. 4/6
I'm labeling Likes and Reposts as Passive Interactions, Replies and Quote posts as Active interactions.
Content-only I need to regenerate the data for, and they're for folks that just post and never interact. 3/6
Content-only I need to regenerate the data for, and they're for folks that just post and never interact. 3/6
May 27, 2025 at 7:48 PM
I'm labeling Likes and Reposts as Passive Interactions, Replies and Quote posts as Active interactions.
Content-only I need to regenerate the data for, and they're for folks that just post and never interact. 3/6
Content-only I need to regenerate the data for, and they're for folks that just post and never interact. 3/6
Remember what I said about Likes? They're gonna skew any of the combined charts for now.
These are all the same ~6-8 hours of data. 46GB on disk.
I do detect and exclude certain user types, but some of them have not been removed from this dataset yet. I wanted a before/after to look at. 2/6
These are all the same ~6-8 hours of data. 46GB on disk.
I do detect and exclude certain user types, but some of them have not been removed from this dataset yet. I wanted a before/after to look at. 2/6
May 27, 2025 at 7:48 PM
Remember what I said about Likes? They're gonna skew any of the combined charts for now.
These are all the same ~6-8 hours of data. 46GB on disk.
I do detect and exclude certain user types, but some of them have not been removed from this dataset yet. I wanted a before/after to look at. 2/6
These are all the same ~6-8 hours of data. 46GB on disk.
I do detect and exclude certain user types, but some of them have not been removed from this dataset yet. I wanted a before/after to look at. 2/6
Ok, there's a HUGE pile of "gotchas" with this data because I need to recollect some things. As an example - likes are all languages, other interacts are just lang=en
#AtProto #BlueskyDev #AtProtoDev #Bluesky
1/6
bluesky-analytics.reskeet.me
#AtProto #BlueskyDev #AtProtoDev #Bluesky
1/6
bluesky-analytics.reskeet.me
May 27, 2025 at 7:48 PM
Ok, there's a HUGE pile of "gotchas" with this data because I need to recollect some things. As an example - likes are all languages, other interacts are just lang=en
#AtProto #BlueskyDev #AtProtoDev #Bluesky
1/6
bluesky-analytics.reskeet.me
#AtProto #BlueskyDev #AtProtoDev #Bluesky
1/6
bluesky-analytics.reskeet.me
WHOO! Solved my memory problem! Process hovering at 6Gb over the last several hours! RAM for days now!
hmm? how? nah, jus.. look at the process list, don--
hmm? how? nah, jus.. look at the process list, don--
May 27, 2025 at 3:33 AM
WHOO! Solved my memory problem! Process hovering at 6Gb over the last several hours! RAM for days now!
hmm? how? nah, jus.. look at the process list, don--
hmm? how? nah, jus.. look at the process list, don--
This is only a snapshot of about an hour, but its similar to longer runs I've seen. The two main users of the network seem to just be passive content producers and passive consumers.
The social users of the "social media platform" are the vast minority... I have some things to think on this....
The social users of the "social media platform" are the vast minority... I have some things to think on this....
May 26, 2025 at 8:15 PM
This is only a snapshot of about an hour, but its similar to longer runs I've seen. The two main users of the network seem to just be passive content producers and passive consumers.
The social users of the "social media platform" are the vast minority... I have some things to think on this....
The social users of the "social media platform" are the vast minority... I have some things to think on this....
Guess I need to find out what @samuel.bsky.team posted 😂
May 26, 2025 at 8:09 PM
Guess I need to find out what @samuel.bsky.team posted 😂
NOOOOO!!! My tool blocked bunnies!!! 😭
I'm sorry bunnies, but I needed to clean up the graph. Your sacrifice is not in vain.
I'm sorry bunnies, but I needed to clean up the graph. Your sacrifice is not in vain.
!["did": "did:plc:3eatnvb2dim417fiwln5wow6",
"reason": "content-producer",
"detectedAt": 1748256847968,
"profile": {
"did": "did:plc:3eatnvb2dim417fiwln5wow6",
"handle": "dailybunnies.bsky. social",
"displayName": "daily bunnies
"followerCount": 62784,
"followsCount": 0,
"labels": [],
"lastUpdated": 1748202761093](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:moykpa7c7np5xyxt4xhdseji/bafkreigqa4lil4voqfzhbe43ua7aa4totnpft3wibg36ykd5xqzz4ijmly@jpeg)
May 26, 2025 at 2:13 PM
NOOOOO!!! My tool blocked bunnies!!! 😭
I'm sorry bunnies, but I needed to clean up the graph. Your sacrifice is not in vain.
I'm sorry bunnies, but I needed to clean up the graph. Your sacrifice is not in vain.
Gentle reminder for the morning that JavaScript / Node.js intervals and timeouts are a *request* for a certain time and not anything close to a guarantee.
My 10 second timeout for stats:
My 10 second timeout for stats:
May 25, 2025 at 3:39 PM
Gentle reminder for the morning that JavaScript / Node.js intervals and timeouts are a *request* for a certain time and not anything close to a guarantee.
My 10 second timeout for stats:
My 10 second timeout for stats:
Oh hey, would you look at that... fighting for first place with Unassuming Local Guy!
May 25, 2025 at 5:25 AM
Oh hey, would you look at that... fighting for first place with Unassuming Local Guy!
Connections are only based on activity, not follows/followers. Density is going to be similar - density of activity. I can adjust the windows for tracking these to get different slices, and some of them I may keep track of for longer and see what I can do with it (have some ideas..) 2/2
May 24, 2025 at 9:20 PM
Connections are only based on activity, not follows/followers. Density is going to be similar - density of activity. I can adjust the windows for tracking these to get different slices, and some of them I may keep track of for longer and see what I can do with it (have some ideas..) 2/2
AHHHHH DATA 😍 Bluesky *live* interaction data - don't take this as canon, this is my first test. lang=en only, and I have 'like' tracking disabled at the moment.
This is so cool to look at. 1/2
This is so cool to look at. 1/2
May 24, 2025 at 9:20 PM
AHHHHH DATA 😍 Bluesky *live* interaction data - don't take this as canon, this is my first test. lang=en only, and I have 'like' tracking disabled at the moment.
This is so cool to look at. 1/2
This is so cool to look at. 1/2
Watching Bluesky data is truly fascinating. I definitely see some recurrent user-types (like folks who primarily repost other ppl's content, content publishers that never interact, etc.)
~3 minutes of data, stats are skewed because I'm only looking at lang=en posts. This is only users who interact:
~3 minutes of data, stats are skewed because I'm only looking at lang=en posts. This is only users who interact:
May 24, 2025 at 3:54 PM
Watching Bluesky data is truly fascinating. I definitely see some recurrent user-types (like folks who primarily repost other ppl's content, content publishers that never interact, etc.)
~3 minutes of data, stats are skewed because I'm only looking at lang=en posts. This is only users who interact:
~3 minutes of data, stats are skewed because I'm only looking at lang=en posts. This is only users who interact: