Skip to main content

Hola, consulté con el ChatGPT y me dice que puede identificar ironía o sarcasmo en lo que le escribo. ¿Esta Brandwatch trabajando en aplicar esta tecnología?, nos dicen que con los custom classifiers se puede hacer, pero es extremadamente difícil hacerlo y que funcione, especialmente en Latam, en donde el español es tan variado en cada país. Gracias por la invitación!

 

Hola @juanca 

How Brandwatch is Building Game-Changing New AI Features Powered by GPT
 

Esto ayuda?


Estaba por responder lo mismo que Ian, al momento para el español habría que ver si esto se aplica bien a la automatización del sentiment y emotions. Sería ideal realizar testings, si alguien tiene un beta ¿me comenta si pudo ver resultados positivos?

 

Muchas gracias


Me parece muy buena la sugerencia De Francisco, ¿Se podrá hacer?


Si, pero no es específicamente lo que consulto sobre sentimient e ironías o sarcasmos, ¿o si? Muchas gracias igualmente

 


Gracias!, lo que me  mandas es super util, pero no especifica si con ChatGPT o similar la plataforma va a poder identificar IRONÍA Y SARCASMO en  las menciones y mejorar asi un problema clave en la evaluación de sentimiento. 

 


Mi conversación con ChatGPT:

As an AI language model, I can certainly attempt to identify irony or sarcasm in a text. However, it's important to note that detecting irony and sarcasm accurately can be challenging even for humans, as they often rely on contextual cues and knowledge about the speaker's intent. In written text, the absence of tone of voice and nonverbal cues makes it more difficult to discern irony or sarcasm accurately. I can make an educated guess based on patterns and keywords, but there may still be instances where my interpretation could be incorrect.

 

could you identify manipulating intentions in a social media post?

 

As an AI language model, I can analyze the language used in a social media post and identify certain patterns or keywords that may suggest manipulating intentions. Some common manipulative tactics used in social media posts include:

  1. Emotional manipulation: This involves using language that plays on the reader's emotions, such as using overly dramatic language or appealing to the reader's sense of pity.

  2. Misinformation or propaganda: This involves spreading false or misleading information in order to sway the reader's opinion or beliefs.

  3. Gaslighting: This involves manipulating the reader into questioning their own perceptions or beliefs, often by denying or distorting facts.

  4. Personal attacks or ad hominem attacks: This involves attacking the character of the reader or another person in order to discredit their opinions or arguments.

  5. Fear-mongering: This involves using language that creates fear or panic in the reader, often by exaggerating or distorting the facts.

By analyzing the language used in a social media post, I can help identify these manipulative tactics and alert the reader to the possibility that the post may be intended to manipulate their opinions or beliefs. However, it's important to note that language is complex, and context is important in determining the true intention behind a post.

 

 

Was this response better or worse?

BetterWorseSame

 

 


Are Brandwatch considering get better results in sentiment analysis whit ChatGPT? . In Spanish we have always to correct many sentiments of the post. I guess in English is better.

 


He estado probando lo nuevo con ChatGPT y es fantástico. Igualmente sigo considerando, aunque no soy técnico, sería ideal para mejorar el algoritmo de sentimiento en cuanto este tiene la limitación de no poder identificar ironía, sarcasmo… doble sentido en general. ChatGPT si puede hacerlo. de hecho, me ha pasado en BCR, que cuando analiza una cantidad x de menciones, en muchos casos identifica que hay menciones con estas características.

Es un problema todavía en el análisis de sentimiento actual, que hay que , o entrenar al sistema con Custom Classifiers, que es limitado o corregir manualmente miles de menciones. Por lo menos en español. Nunca he podido hacer una entrega de un reporte si hablerlo hecho previamente. 

No estoy diciendo que el modelo actual no es bueno, al contrario, creo que con ChatGPT es perfectible. Sería bueno saber cuales son los planes a futuro con esta tecnología

 


Hola @juanca 

 

Mis disculpas, mi español no es lo suficientemente bueno como para discutir inteligencia artificial, así que he pedido a ChatGPT que lo traduzca del inglés por mí.

Es cierto que los grandes modelos de lenguaje podrían ser capaces de proporcionar mejores formas de identificar cosas como el sarcasmo en publicaciones sociales, y sí, estamos explorando cómo se podrían aplicar estos nuevos avances en IA al análisis de sentimiento en Brandwatch.

Sin embargo, el problema en este momento radica en la escala. Por ejemplo, estimamos que ejecutar todos los datos recopilados por Brandwatch a través de ChatGPT costaría decenas de millones de dólares al día. Por lo tanto, en este momento, utilizar ChatGPT u otros modelos similares para realizar análisis de sentimiento en Brandwatch no es una solución viable. No obstante, en el pasado hemos tenido éxito adaptando modelos para trabajar con conjuntos de datos muy grandes, y estamos explorando formas en las que podemos utilizar la nueva tecnología para crear modelos personalizados de sentimiento para diferentes situaciones.

También vale la pena mencionar que nuestro modelo de sentimiento actual maneja el sarcasmo hasta cierto punto. Por lo general, esto depende de qué tan obvio sea en la publicación. Nuestro sentimiento se basa en modelos de arquitectura de transformadores que tienen una comprensión preconstruida de cómo se utiliza el lenguaje. Si está analizando patrones en el texto que son comúnmente vistos en publicaciones sarcásticas o respuestas, puede captar esto e identificar el sentimiento correcto. Pero si el sarcasmo es menos obvio, tendrá dificultades.

Finalmente, también vale la pena recordar que la precisión de la IA en cosas como el sentimiento solo se puede medir en comparación con datos subjetivos. Se ha descubierto que dos humanos solo estarán de acuerdo en el sentimiento de los tweets alrededor del 80% del tiempo. ¡Un día típico en Twitter te mostrará que muchos humanos no pueden detectar el sarcasmo en el texto igual de bien! Por lo tanto, comprender o medir cuán bien puede manejar la IA el sarcasmo es una pregunta complicada.


Hi Ian, also translated by our friend ChatGPT: Great, thank you very much for your thorough explanation. I was only concerned that it wasn't being considered, and it's obvious that the cost would make it impossible. But this is evolving very quickly, and they will surely find a solution. On another note, taking advantage of your willingness, I've always wondered why BW hasn't directly used the Crimson Exagon tool, training it, perhaps country by country (I imagine that's the problem), to learn to recognize irony or sarcasm. Maybe it's not possible, is it? Another topic that may be important that I think I already mentioned: I have seen several platforms that can assess the percentage possibility that a post has been made by a bot or troll. A couple of weeks ago, I was given a demo by an Israeli company, and it was really good. Not only would it be very useful for politics, but it also showed cases of severe attacks against competing companies. What do you think? Thank you very much, and greetings.


I´m sorry!!! Nick, not Ian!!!! Best regards


Hi @juanca, you might be interested in this article...

 


Hi, is Brandwatch considering SORA for BSMM? 


@juanca Hola, not right now - no. But I will add your request to our Product Board.


Reply