Dataset Viewer
Auto-converted to Parquet Duplicate
id
int64
1
9.35k
Tweets
stringlengths
1
310
Description
stringlengths
1
159
1
<link>
always live on the free audacy app
1
are diddy and city girls yung miami a couple <link> via 102jamz
always live on the free audacy app
1
this is sooo real
always live on the free audacy app
1
jayz explains why he beyonce sat during super bowl national anthem <link> via billboard
always live on the free audacy app
1
bring in 2020 wit my dogs djjflex djlilvegas on 102jamz
always live on the free audacy app
1
its time for the morning mix with djlilvegas and 3livecrew tell siri or alexa to play 102 jamz
always live on the free audacy app
1
man yall make a joke out of every damn thing lmaoooo
always live on the free audacy app
1
listen to 3livecrewpodcast
always live on the free audacy app
1
how many disney princesses or disney movies can you name in 30 seconds did osamabindrankin go too far in his ratchetreport big10acc recap w bdaht roxie2hot4tv has a cardi b offset update the 3 live rewind <link> 102jamz
always live on the free audacy app
1
gervonta davis arrested for domestic violence over viral video <link> via tmz
always live on the free audacy app
1
heres the halloween edition of the 3livecrewpodcast listen here and share <link> osamabindrankin roxie2hot4tv 102jamz
always live on the free audacy app
1
thotspiration throwbacks all morning long w 3livecrew turn your radio on rock with us
always live on the free audacy app
1
if you ever miss 3livecrew in the morning you can get your fix by listening to the 3livecrewpodcast
always live on the free audacy app
1
the history of juneteenth from abraham lincoln to today <link> via 102jamz
always live on the free audacy app
1
michael k williams the wire actor found dead in nyc apartment <link> via 102jamz
always live on the free audacy app
1
tory lanez just had his bail increased after this moment with megan thee stallion <link> via 102jamz
always live on the free audacy app
1
vote now for <link> hip hop rb artist of the year <link> via 102jamz
always live on the free audacy app
1
today is the thankful episode w 3livecrew what are you thankful for this morning
always live on the free audacy app
1
3livecrew 102jamz i was dying after that aka response
always live on the free audacy app
1
bdaht 102jamz big sean is so slept on
always live on the free audacy app
1
tune in now to fatjeez187 and djjflex for the 5 oclock remix tell siri or alexa to play 102 jamz
always live on the free audacy app
1
102jamz djlilvegas 3livecrew loving the mix this morning
always live on the free audacy app
1
mtv vmas will return to brooklyn in september <link> via 102jamz
always live on the free audacy app
1
102jamz man i wish my wife was a tarheels fan so all i had to say was carolina suck so i wouldnt have to get her anything for valentine day 3livecrew
always live on the free audacy app
1
3liverewind is up click the link enjoy
always live on the free audacy app
1
talking about this right now free4allfriday <link>
always live on the free audacy app
1
you never have to go a day without 3 live crew if you missed the show this morning click the link and enjoy bdaht roxie2hot4tv osamabindrankin 3 live crew polls the ratchet report is oprah a traitor on soundcloud np <link>
always live on the free audacy app
1
today listeners walked through the secret garden letting us know family members they did not wanna see for thanksgiving bdaht has some things you should watch for tv tuesday on this episode of 3livecrewpodcast <link> 102jamz roxie2hot4tv
always live on the free audacy app
1
lebron on expanding a program for his oldest i promise network kids to include free tuition for four years at kent state its probably the best thing ive ever done via kylegoon
always live on the free audacy app
1
<link>
always live on the free audacy app
1
sorry were so late better late than never tho this mornings show is up click the link give us a listen bdaht osamabindrankin roxie2hot4tv
always live on the free audacy app
1
thegirljt is out and living her best life on her first day out did you listen to her freestyle 102jamz
always live on the free audacy app
1
are you tuned in to the 5 oclock remix with fatjeez187 and djjflex
always live on the free audacy app
1
you missed 3livecrew this morning dont worry we got you w 3liverewind click the link catch up on everything you missed from todays show ohnoshedidnt karol pulled a jussie smollett roxie goes off on jr smith trump talk on soundcloud np <link>
always live on the free audacy app
1
not according to bdaht <link>
always live on the free audacy app
1
are you tuned in fatjeez187 is on your radio right now for the kickback tell siri or alexa to play 102 jamz
always live on the free audacy app
1
let 3livecrew send you to work with a smile on your face get your news gossip games and of course get your laughsturn your radio on well get your morning started right
always live on the free audacy app
1
the cousin probably used her recipe and it was better than her own mac cheese twittercom102jamzstatu
always live on the free audacy app
1
<link>
always live on the free audacy app
1
3liverewind tupac still alive bdaht cries tarheel tears birthday trivia goes off the rails on soundcloud np <link>
always live on the free audacy app
1
mary j blige documentary in the works from amazon entertainment one <link> via thr
always live on the free audacy app
1
tune in now for the 5 oclock remix with djjflex and fatjeez187 tell siri to play 102 jamz
always live on the free audacy app
1
cardi b and offset welcome baby number 2 <link> via 102jamz
always live on the free audacy app
1
the first kanye west and gap collab has arrived and will cost you 200 <link> via 102jamz
always live on the free audacy app
1
are you tuned in with fatjeez187 tell siri or alexa to play 102 jamz
always live on the free audacy app
1
always live on the free audacy app
always live on the free audacy app
1
f4af fruit roll up challengeis it for couples only madness in the nfl lil baby drug rumors key key sings for the triaddid we like it listen to find out have you heard free 4 all freaking friday by 3 live rewind on soundcloud np <link>
always live on the free audacy app
1
oopyou think bey is being shady big orange bloop beyonce apparently sent everybody but kim k an iconic ivy park box
always live on the free audacy app
1
honestly because why would someone bring 52 stop a dude from getting some yams
always live on the free audacy app
1
get your daily dose of 3livecrew listen share
always live on the free audacy app
1
chile jerry dun said r kellys song was fiasco fiasco i cant please have him call back in we need more from him 3livecrew 3livecrew 102jamz
always live on the free audacy app
1
hey 102jamz can you let roxie2hot4tv see this twittercomstfuanajaist
always live on the free audacy app
1
102jamz yall is definitely the algorithm to waking up right
always live on the free audacy app
1
thotspiration throwbacks trump talk news gossip major laughs all morning long get your thursday morning started w 3livecrew
always live on the free audacy app
1
bobbi kristinas bf nick gordon dies from suspected drug overdose <link> via tmz
always live on the free audacy app
1
we knew yo ass would be back ya boys looked good
always live on the free audacy app
1
did you miss 3livecrew this morning click the link and listen
always live on the free audacy app
1
102jamz for the quick mix all the artist name start with d roxie2hot4tv bdaht osamabindrankin
always live on the free audacy app
1
8 lies i tell my husband that make our marriage 1000 stronger 1 the sex is mind blowing 2 no you havent gained weight 3 i still think youre hot 4 i loved the gift you gave me get the last 4 in an hour on 102jamz w 3livecrew
always live on the free audacy app
1
whatever they say he didhe did that ish <link>
always live on the free audacy app
1
appreciate you man 3livecrew <link>
always live on the free audacy app
1
its time for the morning mix with djlilvegas and 3livecrew tell siri or alexa to play 102 jamz now
always live on the free audacy app
1
we find out how much time tekashi gets on 1218 tekashi 6ix9ine prosecutors file request for lighter sentence <link> via tmz
always live on the free audacy app
1
next wednesday december 11th 102jamz alongside nse join forces for the no stress charity bowl we ask that everyone that attends bring an unwrapped gift the following day we will be dropping off at the <link>
always live on the free audacy app
1
shaq lost a bet to dwade now america knows what his hairline looks like via nbaontnt
always live on the free audacy app
1
visit <link> for your chance to win your way into the fsoladiesfirst virtual womens empowerment event with iyanlavanzant presented by ambetter of north carolina inc
always live on the free audacy app
1
congrats nickiminaj
always live on the free audacy app
1
register now for hbcu college day tour presented by 3815 media on saturday september 18th 10am6pm at bowman gray stadium west parking lot admission is free but you must register here <link> hbcuawardsshow
always live on the free audacy app
1
tune in on the <link> app
always live on the free audacy app
1
random radio wednesday w the crew <link> via audioboom
always live on the free audacy app
1
3livecrew wants to talk to you this morning what was the worst non violent punishment you ever received as a child 336 6801021
always live on the free audacy app
1
3rd row tix to ncat homecoming concert are up for grabs w 3livecrew this morning all you have to do is win the game lip service
always live on the free audacy app
1
listen to the show w no interruptionsno music no commercials all the funny click the link and enjoy 3liverewind is up now twittercomdrankontherox
always live on the free audacy app
1
somebody is trolling julia roberts was suggested to play harriet tubman by studio exec says harriet screenwriter <link>
always live on the free audacy app
1
wonder if djlilvegas gone do a jayz mix at 9 102jamz bdaht roxie2hot4tv osamabindrankin
always live on the free audacy app
1
the rain is coming this evening heres a list of indoor or rescheduled halloween events
always live on the free audacy app
1
sitting at work in the car eating my breakfast tunned in 3livecrew yhall get my day going 102jamz i thank yhall
always live on the free audacy app
1
lori harvey cited for hit and run released onsite <link> via tmz
always live on the free audacy app
1
fort worth officer opens fire killing woman inside her home pd <link> via nbcdfw
always live on the free audacy app
1
girlwtf you think was going to happen
always live on the free audacy app
1
3livecrew is on your radio right now and we want to put a smile on your face to start your day turn us on and get your news gossip and laughs
always live on the free audacy app
1
are rappers glorifying drug use the cause of juiceworld a death sundaytea missperry92 sheismonasia
always live on the free audacy app
1
i been low key missing home for two weeks so i stream 102jamz 3livecrew bdaht roxie2hot4tv osamabindrankin wayyy in atl to get a good laugh in whew i miss nc
always live on the free audacy app
1
rihanna shows off that new new in cheeky wrapped in ribbon lingerie instagram post <link> via 102jamz
always live on the free audacy app
1
caught up in the quick mix with djlilvegas thanks for getting lit before clocking it 102jamz 3livecrew
always live on the free audacy app
1
soo i was a fat ass 1st graderone day i was the only kid answering questions and my teacher said your on it today malik if i had another lunch id give it to you twittercomtransvampires
always live on the free audacy app
1
good morninggggggg its the 1st free 4 all friday of 2020 3livecrew is letting you control the radio call up get the conversation started 336 6801201
always live on the free audacy app
1
3liverewind
always live on the free audacy app
1
dababydababy s baby
always live on the free audacy app
1
<link>
always live on the free audacy app
1
cleevon just walked in the studio and hes ready to give you your thotspiration turn ya radio on and rock w 3livecrew
always live on the free audacy app
1
so if its not her who is it and where is she signs r kellys exgf joycelyn might not be behind tellall posts <link> via tmz
always live on the free audacy app
1
iggy didnt like these comments from ti ti admits hes trying to undo the blunder of iggy azalea <link>
always live on the free audacy app
1
visit <link> for your chance to win your way into the fsoladiesfirst virtual womens empowerment event with iyanlavanzant presented by ambetter of north carolina inc
always live on the free audacy app
1
cleevons car broke down this morning so osamabindrankin had to pick him up for work but he made it hes ready to give you your weekly dose of thotspiration
always live on the free audacy app
1
wait a minuteunc fans dont consider duke their rival old names are making a comeback the m <link> via audioboom
always live on the free audacy app
1
vote now for <link> hip hop rb artist of the year <link> via 102jamz
always live on the free audacy app
1
us aka her sorority us <link>
always live on the free audacy app
1
3livecrew is taking suggestions for donald trumps 2020 campaignhere are some ideas weve gotten so far 1 impeachment was the case that they gave me 2 make america great again again 3 surviving impeachment what do you think his slogan will be
always live on the free audacy app
1
3livecrew is your hook up for those 5th row ghoe concert tickets <link>
always live on the free audacy app
End of preview. Expand in Data Studio

Dataset Card for Twitter Authorship Verification Dataset

Dataset de pares de tuits (o grupos de tuits) en inglés, etiquetados según si pertenecen al mismo usuario o a usuarios distintos, construido para entrenar y evaluar modelos de verificación de autoría en Twitter. Fue creado como parte del Trabajo de Fin de Grado "Aprendizaje automático para la detección de autoría de usuarios en Twitter" (Universidad de Murcia, 2023).

Dataset Details

Dataset Description

El dataset se construyó recolectando tuits de una muestra de usuarios de Twitter en inglés y transformándolos en un corpus de pares (o grupos) de contribuciones etiquetadas con 1 si pertenecen al mismo usuario y 0 si pertenecen a usuarios distintos. El objetivo es entrenar modelos capaces de discernir autoría entre dos contribuciones textuales, en lugar de clasificar directamente qué usuario escribió un texto (verificación de autoría, no identificación cerrada de autor).

El proceso de construcción sigue cuatro fases: obtención de datos (web scraping), filtrado (idioma y contenido), vectorización (bolsa de palabras o afinidad temática) y emparejamiento (generación de pares mismo/distinto usuario, con opción de balanceo o desbalanceo).

  • Curated by: Carlos Andrés Vázquez Vicente
  • Funded by: No aplica (proyecto académico, TFG)
  • Shared by: Carlos Andrés Vázquez Vicente
  • Language(s) (NLP): Inglés (en)
  • License: No especificada en el TFG original — a determinar por el autor antes de publicar.

Dataset Sources

  • Repository: No especificado (dataset generado localmente durante el TFG, no publicado previamente en un repositorio)
  • Paper: Vázquez Vicente, C. A. (2023). Aprendizaje automático para la detección de autoría de usuarios en Twitter (Trabajo de Fin de Grado). Universidad de Murcia.
  • Demo: No disponible

Uses

Direct Use

Entrenamiento y evaluación de modelos de verificación de autoría (mismo autor / distinto autor) sobre texto corto en inglés estilo microblog, incluyendo:

  • Modelos clásicos de clasificación vectorial (p. ej. SVC) sobre representaciones de bolsa de palabras (BOW) o afinidad temática.
  • Fine-tuning de modelos de lenguaje basados en transformers (BERT, DistilBERT, XLM-RoBERTa) para la tarea de clasificación binaria de pares de texto.
  • Investigación académica sobre el efecto de la agrupación de tuits por usuario, el número de tokens y el tipo de vectorización en el rendimiento de estos modelos.

Out-of-Scope Use

No debe usarse como herramienta única o determinante para identificar, perseguir, doxear o tomar acciones legales/disciplinarias contra personas reales basándose en la autoría predicha. El propio TFG concluye que estos modelos deben usarse como apoyo a estudios más amplios, no como prueba autónoma de autoría. Tampoco está pensado para perfilado de usuarios en producción, detección de bots (no evaluado para ese fin) ni para idiomas distintos del inglés.

Dataset Structure

El dataset final es un dataframe/CSV donde cada fila representa un par de entradas de texto (tuit individual o grupo concatenado de tuits) y una etiqueta binaria:

  • texto_1: contenido textual (tuit o concatenación de N tuits) del primer usuario del par.
  • texto_2: contenido textual (tuit o concatenación de N tuits) del segundo usuario del par.
  • label: 1 si ambos textos pertenecen al mismo usuario, 0 si pertenecen a usuarios distintos.

Existen variantes del dataset según:

  • Nivel de agrupación: sin agrupar (1 tuit), o agrupaciones de 2, 5, 10 y 25 tuits por entrada.
  • Vectorización: representación por bolsa de palabras (BOW, con distintos tamaños de vocabulario: 50 a 3000 tokens) o representación por afinidad a 19 temas (usando el modelo tweet-topic-21-multi de Cardiff NLP).
  • Balanceo: por defecto 50% pares del mismo usuario / 50% pares de usuarios distintos, con funciones auxiliares para generar versiones desbalanceadas.

No se documentan splits oficiales de train/validation/test en el TFG; estos se generaron ad hoc en cada experimento.

Dataset Creation

Curation Rationale

No existía en el momento del TFG un dataset público adecuado para la tarea específica de verificación de autoría en Twitter (a diferencia de tareas más estudiadas como el análisis de sentimiento). El dataset se generó desde cero para cubrir ese vacío, priorizando el idioma inglés (mayoritario en Twitter y en modelos preentrenados de NLP) y tratando de minimizar sesgos en la selección de usuarios.

Source Data

Data Collection and Processing

Los datos se obtuvieron mediante web scraping con BeautifulSoup y Selenium sobre perfiles públicos de Twitter, sin sesión iniciada (para evitar sesgos de personalización). Los usuarios se obtuvieron por dos vías: (1) búsqueda de preposiciones (palabras sin carga temática) en Twitter sin sesión activa, para diversificar temáticas, y (2) una lista de usuarios verificados del proyecto "TimeNLP" de la Universidad de Cardiff. Se recolectaron entre 50 y 150 tuits por cuenta.

Procesamiento posterior:

  • Filtro de idioma: detección de idioma con spaCy (en_core_web_sm), reduciendo el conjunto a ~10.000 usuarios en inglés, más una revisión manual adicional.
  • Filtro de contenido: eliminación de emojis y caracteres extraños, texto en minúsculas, sustitución de hipervínculos por el token link, y eliminación de cuentas cuyo contenido quedó demasiado reducido tras el filtrado.
  • Vectorización: bolsa de palabras (filtrando previamente por sustantivos/verbos/adjetivos con spaCy y por frecuencia) o afinidad temática con el modelo tweet-topic-21-multi (Cardiff NLP, basado en RoBERTa).
  • Emparejamiento: script en Python que agrupa tuits por usuario, repite y empareja filas asignando un id secundario (mismo usuario 50% de las veces, distinto 50%), generando la etiqueta binaria final.

Who are the source data producers?

Usuarios públicos de Twitter (ahora X) que publican en inglés, seleccionados mediante búsqueda de palabras neutrales (preposiciones) sin sesión iniciada, y usuarios verificados listados por el proyecto "TimeNLP" de Cardiff NLP. No se dispone de información demográfica autoinformada de estos usuarios; no se realizó anonimización explícita más allá de la eliminación de emojis y caracteres extraños durante el filtrado.

Annotations

Annotation process

No hay anotación manual de autoría: la etiqueta (mismo/distinto usuario) se deriva automáticamente y de forma determinista a partir de los identificadores de usuario ya conocidos durante la recolección (no es un juicio humano subjetivo, sino un emparejamiento programático).

Who are the annotators?

Generadas de forma automática por el script de emparejamiento desarrollado por el autor del TFG; no hubo anotadores humanos.

Personal and Sensitive Information

El dataset contiene texto de tuits públicos, que puede incluir opiniones personales, referencias a terceros, ubicaciones o cualquier contenido que el usuario decidiera publicar públicamente. No se realizó anonimización de nombres de usuario, menciones ni contenido más allá de la eliminación de emojis/caracteres extraños y la sustitución de URLs por el token link. Al provenir de contenido público de redes sociales, existe riesgo de identificabilidad de los autores originales; se recomienda revisar cuestiones de privacidad y consentimiento antes de una publicación pública del dataset.

Bias, Risks, and Limitations

  • Los usuarios de redes sociales no son representativos de la población general y pueden introducir sesgos demográficos, temáticos o lingüísticos (mencionado explícitamente en el TFG, citando a Leng, 2013).
  • El método de selección de usuarios (búsqueda de preposiciones + lista de verificados de Cardiff NLP) puede sesgar hacia cuentas más activas, verificadas o con presencia pública notable.
  • El tamaño del dataset y la agrupación de tuits afectan fuertemente el rendimiento de los modelos; agrupaciones mayores mejoran la precisión pero reducen el tamaño efectivo del conjunto y aumentan el riesgo de redundancia si se usa repetición combinatoria.
  • Restringido a texto en inglés; no se ha validado su aplicabilidad a otros idiomas.
  • No evaluado en escenarios reales como suplantación de identidad, cambios de titularidad de cuentas o detección de bots.

Recommendations

Los usuarios de este dataset deben ser conscientes de los sesgos de selección de usuarios y del origen público (no anonimizado) del contenido. Se recomienda no utilizar el dataset ni los modelos resultantes como única evidencia de autoría en contextos legales o disciplinarios, y considerar ampliar la recolección de datos y aplicar métodos de anonimización más robustos antes de cualquier publicación o uso a mayor escala.

Citation

APA:

Vázquez Vicente, C. A. (2023). Aprendizaje automático para la detección de autoría de usuarios en Twitter [Trabajo de Fin de Grado, Universidad de Murcia].

BibTeX:

@thesis{vazquezvicente2023autoria,
  title        = {Aprendizaje automático para la detección de autoría de usuarios en Twitter},
  author       = {Vázquez Vicente, Carlos Andrés},
  school       = {Universidad de Murcia},
  year         = {2023},
  type         = {Trabajo de Fin de Grado}
}

Glossary

  • BOW (Bag of Words): representación de texto como conjunto desordenado de palabras según su frecuencia.
  • Emparejamiento: proceso de combinar tuits o grupos de tuits de distintos usuarios (o del mismo) para generar la etiqueta de autoría.
  • Verificación de autoría: tarea de determinar si dos textos fueron escritos por la misma persona, a diferencia de la identificación de autoría (asignar un texto a un autor concreto de un conjunto cerrado).

More Information

Para más detalle sobre el proceso de construcción del dataset, metodología, resultados de los modelos (SVC, DistilBERT, BERT base, XLM-RoBERTa) y limitaciones, consultar el TFG completo: Aprendizaje automático para la detección de autoría de usuarios en Twitter (Universidad de Murcia, 2023), especialmente el capítulo 4 ("Diseño y resolución").

Dataset Card Authors

Carlos Andrés Vázquez Vicente (con asistencia de Claude para la redacción del card a partir del TFG).

Dataset Card Contact

carlosandres.vazquezv@um.es

Downloads last month
27