Artificial Intelligence Sentiment Analysis. If “I like to kill kitties”, would it be positive ?
Views (547)
Jalmaraz
669
Like
Share
ReportArtificial Intelligence (AI) Services overview.
AI services generally process a human like input (natural language text, image or sound) and returns a serializable analysis of this human like input. Output usually is a JSON stream. This will not be a very technical post, but this is the general review about how a program can manage an AI service:
We have to subscribe to the service, get an API key, and an API URL. With these credentials your program can access to the service.
- Take the human like input (Natural language text).
- Assembly a HTTP request with this input to the AI service.
- Receive the JSON response and handle the data, making checks or storing it in a database.
- The final purpose is batching this process to analyze lots of texts.
You can see an AI API call example in this post: https://community.dynamics.com/nav/b/conceptsindailynavisionwork/archive/2019/02/17/cognitive-entity-api-services-with-al-did-you-mean .
Sentiment analysis.
Sentiment analysis is an Artificial Intelligence service, that extracts affective state from a natural language text. So, the input is a Natural text “I don´t like drink tea”, and returns this JSON:
{
"sentiment": {
"document": {
"score": -0.562822,
"label": "negative"
}
}
}
The output takes one of these values: negative, positive or neutral.
A common use case is analyzing a brand (company, item) perception from opinions of users. Basically, we get an opinion analysis, from the author point of view.
Service suppliers.
There are companies that offer API sentiment analysis services. That´s a personal overview from a newcomer (Viñetas):
Microsoft: I like it for its simplicity. It´s very friendly for developers. We can do a manual demo in link https://azure.microsoft.com/es-es/services/cognitive-services/text-analytics/ .
Google: I like it for its very deep analysis capabilities, even in Spanish languages: gets stylistic nuances and circumlocution and believe in Spanish is not an easy job. You play with this in https://cloud.google.com/natural-language/?hl=es
Sentiment analysis paradoxes. The tricky denomination.
I like to test with strange texts or images these services to learn more about its behavior. So, let´s come in in Microsoft and type this text:
Positive output (opinion green color and 85 % is very positive) could be madness for many people who talked before about this, and they feel this like a bad performance of the tool. The idea of the post begins when someone deeply shocked told me this.
But we are talking about sentiment analysis, not moral or psychiatric analysis. Not about is good or bad this action, it´s about the (insane in this example) feelings of author and he likes to do something. I think in the branding perception handling this is the right analysis approach.
This make a lot of sense if we use the other name of Sentiment Analysis: Opinion Mining. “Opinion Mining” is a very much accurate than “Sentiment analysis”, and could be less confusing.
Spanish subtle expression.
These services are improving every day. If I type this in Spanish “Deja mucho que desear la calidad del servicio”, I mean “There is a lack of service quality”, but in a subtle way. Many weeks ago, Cognitive didn´t understood this circumlocution, and detect it as positive opinion, focusing in words “quality services”.
You only can believe me right now about that because, currently if I type this again, it detects it as a negative opinion. So, we have continuously improvement in its knowledge database:
*This post is locked for comments