Lmst

Oscar Kjell boosted:

The {text} #rstats 📦 can create word embeddings “ready to be used for downstream tasks. [It] provides a user-friendly link to language models based on transformers from Hugging Face.”
{text} also aims to be “an end-to-end solution that provides state-of-the-art AI techniques tailored for social and behavioral scientists.” There are functions to compute semantic similarities, analyze language, predict text, plot words & more.
By @oscarkjell
https://r-text.org
#NLP #AI #GenAI

The textEmbed() function is the main embedding function in text; and can output contextualized embeddings for tokens (i.e., the embeddings for each single word instance of each text) and texts (i.e., single embeddings per text taken from aggregating all token embeddings of the text).

library(text)
# Transform the text data to BERT word embeddings

# Example text
texts <- c("I feel great!")

# Defaults
embeddings <- textEmbed(texts)
embeddings
See Get Started for more information.

Language Analysis Tasks
It is also possible to access many language analysis tasks such as textClassify(), textGeneration(), and textTranslate().

Semantic similarities and distances
textSimilarity()

Compute the semantic similarity between two text variables.

textDistance()

Compute the semantic distance between two text variables.

textSimilarityMatrix()

Compute semantic similarity scores between all combinations in a word embedding

textDistanceMatrix()

Compute semantic distance scores between all combinations in a word embedding

textSimilarityNorm()

Compute the semantic similarity between a text variable and a word norm (i.e., a text represented by one word embedding that represent a construct).

textDistanceNorm()

Compute the semantic distance between a text variable and a word norm (i.e., a text represented by one word embedding that represent a construct/concept).

Language Analysis Tasks
textClassify()

Predict label and probability of a text using a pretrained classifier language model. (experimental)

textGeneration()

Predicts the words that will follow a specified text prompt. (experimental)

textNER()

Named Entity Recognition. (experimental)

textSum()

Summarize texts. (experimental)

textQA()

Question Answering. (experimental)

textTranslate()

Translation. (experimental)

textZeroShot()

Zero Shot Classification (Experimental)

Oscar Kjell boosted:

the #Rtext tutorial is now published in #PsychologicalMethods

the 📦 1.0 is finally available from #CRAN

if u update, also re-install the conda environment with textrpp_install()

https://psyarxiv.com/293kt
#rstats #NLP #ML
#MachineLearning

the #Rtext tutorial is now published in #PsychologicalMethods

the 📦 1.0 is finally available from #CRAN

if u update, also re-install the conda environment with textrpp_install()

https://psyarxiv.com/293kt
#rstats #NLP #ML
#MachineLearning

Oscar Kjell boosted:

The second edition of R for Data Science is out now, and it’s been a blast to join @hadleywickham as a co-author! Learn more about the second edition at https://www.tidyverse.org/blog/2023/07/r4ds-2e/, read it for free at https://r4ds.hadley.nz/, or buy a copy at https://amzn.to/3PTdLRQ. #rstats

Oscar Kjell boosted:

This is the official Mastodon account of the useR! conference.

Overseen by @R_Foundation, useR! is the main meeting of the global R user and developer community.

It features both invited and user-contributed presentations, as well as tutorials, posters, and other sessions.

Follow us for the latest news on the upcoming conference and the wider conference series!

#RStats

Oscar Kjell boosted:

Shiny for Python is now generally available! 🎉

With Shiny, you can create interactive web apps without HTML/CSS.

Now, you can combine Shiny's reactivity with #Python's vast resources for endless possibilities.

Check out new #PyShiny resources and learn more: https://posit.co/blog/shiny-for-python-is-now-generally-available/

Oscar Kjell boosted:

Some transparency as I am editing a podcast episode of @NulliusInVerba on publication bias. Although we have released all recordings, there is some selective reporting! We edit the episode for a smoother listener experience and to keep episodes around 1 hour. Every grey line is an edit - but you won't notice most of them in the final episode.

Audition software with 2 sound tracks, heavily edited

Oscar Kjell boosted:

@oscarkjell
LLMs as the "the missing piece for natural language responses to replace closed-ended rating scales." That is indeed a tantalizing possibility.

Thanks for posting the paper. Your earlier work (as referenced in the paper) looks really interesting as well.
@handyschwartz @katarinakjell

Oscar Kjell boosted:

CALL FOR PAPERS

#PsychotherapyResearch is inviting submissions for a special section on

"Therapist Responsiveness"

Submissions are continuously accepted until the deadline
🗓️ 31.08.2023
https://mc.manuscriptcentral.com/tpsr

We are looking for contributions supporting the consolidation and integration of knowledge and findings in this emerging area.

Contact the Special Section Co-Editors for more details or with any questions you may have about the call!

#PatientCentered

Screenshot of some key information about the call "Therapist Responsiveness: Developing an empirical basis for the future"

Adaptations of the therapeutic process in response to client expression and behavior have been summarized under the term "therapist responsiveness", describing the act of adapting or the degree to which a therapist (or a setting) is inclined to make such adaptations.

Therapist responsiveness may operate within the therapeutic relationship, as adaptation of a specific therapeutic technique or as an explicit orientation towards client characteristics (including their minority status), or any other instance in the psychotherapy process.

This Special Section of Psychotherapy Research seeks original contributions which should lay the groundwork for the future development of a consolidated and integrative knowledge base on therapist responsiveness. We seek a) empirical studies testing aspects of therapist responsiveness, b) meta-analytic, meta-synthetic or systematic reviews on therapist responsiveness, c) qualitative and quantitative studies in the emerging field of therapist responsiveness.

A particular point to be addressed in any submission is the way the methodology established that the therapist originally intended a different course of action and then adapted their plans or actions.

Special Section Co-Editors:

Ueli Kramer, University of Lausanne

Giovanna Esposito, Università degli Studi di Napoli Federico II

Jan R. Boehnke, University of Dundee

Oscar Kjell boosted:

Moving Experimental Psychology Online: How to Maintain Data Quality When We Can't See Our Participants http://osf.io/2fhcb/

Oscar Kjell boosted:

As a #PeerReview-er / editor of and researcher conducting prospective randomised studies, I remain perplexed by how many registries do not have mandatory fields for

1) Outcome definitions and operationalisations

2) Sample size justifications.

These are among the most important points in the #StudyDesign phase and key motivations of conducting a specific trial.

You could argue that this would be part of the protocol, but: how many trials do have unpublished protocols?

#RCT #Registration

Oscar Kjell boosted:

OK, :fediverse: #Fediverse hive-mind. I'm working on #TwitterMigration research on higher reach individuals on Twitter that have already migrated to the Fediverse, and have not left #Twitter... at least not yet.

Am looking for accounts on Twitter that have about 500K followers and higher and are also here.

▶️ Am I missing any accounts?

Boost this around to help share it, want to make this as complete as I can. Thanks!

Here is who I have so far: 👇

https://docs.google.com/spreadsheets/d/1MdxfIYcO-T6ghtJUXGgvOJ-EmO0eXLWpfNTbSEH50d4/edit?usp=sharing

Oscar Kjell boosted:

{ggsound} first prototype

ggplot(mtcars) +
geom_point(aes(mpg, wt)) +
theme_sound_basic()

#RStats #soundon

Oscar Kjell boosted:

Super proud/excited to share our paper "Pygmalion Displacement: When Humanising AI Dehumanises Women" wherein we develop a lens to help us trace a type of harm towards women within/by AI as a field & as a technology: https://osf.io/preprints/socarxiv/jqxb6

1/4

The series of questions that comprise the Pygmalion lens. For a given technosocial relationship between AI and people, if one or more of the answers are "Yes", then we can conclude that (an aspect of) Pygmalion displacement is occurring, which is damaging to women, and the feminised broadly construed. If not, then the lens does not apply, and we remain agnostic as to gendered or other harm within this framework. Our proverbial lens should be taken inter alia to be akin to an optical lens which allows for the eye to see more details than when naked, as depicted in the line drawing below.

This figure depicts, in the style of a printed circuit board, a timeline of events (many with related images) that involve Pygmalion displacement in one form or another. On the left in pink, fictional instances; and on the right in green, historical individuals or artefacts.

Pygmalion and Galatea; painting by the French artist Jean-Léon Gérôme, c. 1890. The reader is invited to notice the immediate sexual nature of the sculpture and scene generally; and to attend to the fact that even Galatea's upper body, which seems to be coming to life, is depicted as particularly paler than that of Pygmalion. This is in accordance with certain renditions of the myth, in which she is sculpted from ivory.

Oscar Kjell boosted:

As some here know, I was always a huge fan of Twitter 1.0. Even at its worst, and even while hating so many things about it, it was still an amazing invention I passionately evangelized for years. But I was always in love with the concept, more than than its often terrible implementations. I have never lost that love, and I now long to see it rekindled here in the verse where we can build the kinds of knowledge sharing and processing apps that would never happen on Twitter, even at its peak.

Check out this awesome #DanceYourPhD 🕺 video about using language-based psychological assessments #AI #LargeLanguageModels

https://www.youtube.com/watch?v=A50Q3t4TIvw

Oscar Kjell boosted:

If you're :rstats: user and interested in 🤗 #transformers, have a look at the text package by @oscarkjell et al.:

👉 https://r-text.org/

Great documentation, including preprint tutorial paper and example data on OSF.

#rstats #Rtext #HuggingFace #TextAsData #NLP

Oscar Kjell boosted:

I think people have been gate keeping Mastodon a little bit. The narrative that it’s for nerds or difficult to understand or set up seems far fetched.

People understand email and sign up for it right? We used to copy and paste code to add music widgets and confetti to our MySpace page, and that was widely popular.

So why is selecting an instance and understanding the separation of feeds such an ask?

Maybe we should change the conversation around the Fediverse and tell people what it allows, and what it does different than focus on the five minutes of mildly inconvenient setup.

Oscar Kjell boosted:

This thread explains the {quartodon} R 📦 (see https://quartodon.opens.science).

The #rstats quartodon 📦 allows you to post a Mastodon thread from a plain text file (e.g., a blog post from a Quarto, {blogdown}, or {distill} website, another Quarto or R Markdown file, or just a plain text file).

This effectively allows you to post blog posts to Mastodon in a thread of toots 📑➡️🪄➡️🐘🐘🐘

🧵1/10

@matherion @handyschwartz @katarinakjell thanks 😊

Client Info